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 | 13 | #include "mds_client.h" |
99ccbd22 | 14 | #include "cache.h" |
3d14c5d2 | 15 | #include <linux/ceph/osd_client.h> |
1d3576fd SW |
16 | |
17 | /* | |
18 | * Ceph address space ops. | |
19 | * | |
20 | * There are a few funny things going on here. | |
21 | * | |
22 | * The page->private field is used to reference a struct | |
23 | * ceph_snap_context for _every_ dirty page. This indicates which | |
24 | * snapshot the page was logically dirtied in, and thus which snap | |
25 | * context needs to be associated with the osd write during writeback. | |
26 | * | |
27 | * Similarly, struct ceph_inode_info maintains a set of counters to | |
25985edc | 28 | * count dirty pages on the inode. In the absence of snapshots, |
1d3576fd SW |
29 | * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count. |
30 | * | |
31 | * When a snapshot is taken (that is, when the client receives | |
32 | * notification that a snapshot was taken), each inode with caps and | |
33 | * with dirty pages (dirty pages implies there is a cap) gets a new | |
34 | * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending | |
35 | * order, new snaps go to the tail). The i_wrbuffer_ref_head count is | |
36 | * moved to capsnap->dirty. (Unless a sync write is currently in | |
37 | * progress. In that case, the capsnap is said to be "pending", new | |
38 | * writes cannot start, and the capsnap isn't "finalized" until the | |
39 | * write completes (or fails) and a final size/mtime for the inode for | |
40 | * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0. | |
41 | * | |
42 | * On writeback, we must submit writes to the osd IN SNAP ORDER. So, | |
43 | * we look for the first capsnap in i_cap_snaps and write out pages in | |
44 | * that snap context _only_. Then we move on to the next capsnap, | |
45 | * eventually reaching the "live" or "head" context (i.e., pages that | |
46 | * are not yet snapped) and are writing the most recently dirtied | |
47 | * pages. | |
48 | * | |
49 | * Invalidate and so forth must take care to ensure the dirty page | |
50 | * accounting is preserved. | |
51 | */ | |
52 | ||
2baba250 YS |
53 | #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10)) |
54 | #define CONGESTION_OFF_THRESH(congestion_kb) \ | |
55 | (CONGESTION_ON_THRESH(congestion_kb) - \ | |
56 | (CONGESTION_ON_THRESH(congestion_kb) >> 2)) | |
57 | ||
61600ef8 YZ |
58 | static inline struct ceph_snap_context *page_snap_context(struct page *page) |
59 | { | |
60 | if (PagePrivate(page)) | |
61 | return (void *)page->private; | |
62 | return NULL; | |
63 | } | |
1d3576fd SW |
64 | |
65 | /* | |
66 | * Dirty a page. Optimistically adjust accounting, on the assumption | |
67 | * that we won't race with invalidate. If we do, readjust. | |
68 | */ | |
69 | static int ceph_set_page_dirty(struct page *page) | |
70 | { | |
71 | struct address_space *mapping = page->mapping; | |
72 | struct inode *inode; | |
73 | struct ceph_inode_info *ci; | |
1d3576fd | 74 | struct ceph_snap_context *snapc; |
7d6e1f54 | 75 | int ret; |
1d3576fd SW |
76 | |
77 | if (unlikely(!mapping)) | |
78 | return !TestSetPageDirty(page); | |
79 | ||
7d6e1f54 | 80 | if (PageDirty(page)) { |
1d3576fd SW |
81 | dout("%p set_page_dirty %p idx %lu -- already dirty\n", |
82 | mapping->host, page, page->index); | |
7d6e1f54 | 83 | BUG_ON(!PagePrivate(page)); |
1d3576fd SW |
84 | return 0; |
85 | } | |
86 | ||
87 | inode = mapping->host; | |
88 | ci = ceph_inode(inode); | |
89 | ||
90 | /* | |
91 | * Note that we're grabbing a snapc ref here without holding | |
92 | * any locks! | |
93 | */ | |
94 | snapc = ceph_get_snap_context(ci->i_snap_realm->cached_context); | |
95 | ||
96 | /* dirty the head */ | |
be655596 | 97 | spin_lock(&ci->i_ceph_lock); |
7d8cb26d | 98 | if (ci->i_head_snapc == NULL) |
1d3576fd SW |
99 | ci->i_head_snapc = ceph_get_snap_context(snapc); |
100 | ++ci->i_wrbuffer_ref_head; | |
101 | if (ci->i_wrbuffer_ref == 0) | |
0444d76a | 102 | ihold(inode); |
1d3576fd SW |
103 | ++ci->i_wrbuffer_ref; |
104 | dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d " | |
105 | "snapc %p seq %lld (%d snaps)\n", | |
106 | mapping->host, page, page->index, | |
107 | ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1, | |
108 | ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, | |
109 | snapc, snapc->seq, snapc->num_snaps); | |
be655596 | 110 | spin_unlock(&ci->i_ceph_lock); |
1d3576fd | 111 | |
7d6e1f54 SZ |
112 | /* |
113 | * Reference snap context in page->private. Also set | |
114 | * PagePrivate so that we get invalidatepage callback. | |
115 | */ | |
116 | BUG_ON(PagePrivate(page)); | |
117 | page->private = (unsigned long)snapc; | |
118 | SetPagePrivate(page); | |
1d3576fd | 119 | |
7d6e1f54 SZ |
120 | ret = __set_page_dirty_nobuffers(page); |
121 | WARN_ON(!PageLocked(page)); | |
122 | WARN_ON(!page->mapping); | |
1d3576fd | 123 | |
7d6e1f54 | 124 | return ret; |
1d3576fd SW |
125 | } |
126 | ||
127 | /* | |
128 | * If we are truncating the full page (i.e. offset == 0), adjust the | |
129 | * dirty page counters appropriately. Only called if there is private | |
130 | * data on the page. | |
131 | */ | |
d47992f8 LC |
132 | static void ceph_invalidatepage(struct page *page, unsigned int offset, |
133 | unsigned int length) | |
1d3576fd | 134 | { |
4ce1e9ad | 135 | struct inode *inode; |
1d3576fd | 136 | struct ceph_inode_info *ci; |
61600ef8 | 137 | struct ceph_snap_context *snapc = page_snap_context(page); |
1d3576fd | 138 | |
4ce1e9ad | 139 | inode = page->mapping->host; |
b150f5c1 MT |
140 | ci = ceph_inode(inode); |
141 | ||
142 | if (offset != 0 || length != PAGE_CACHE_SIZE) { | |
143 | dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n", | |
144 | inode, page, page->index, offset, length); | |
145 | return; | |
146 | } | |
4ce1e9ad | 147 | |
99ccbd22 MT |
148 | ceph_invalidate_fscache_page(inode, page); |
149 | ||
150 | if (!PagePrivate(page)) | |
151 | return; | |
152 | ||
1d3576fd SW |
153 | /* |
154 | * We can get non-dirty pages here due to races between | |
155 | * set_page_dirty and truncate_complete_page; just spit out a | |
156 | * warning, in case we end up with accounting problems later. | |
157 | */ | |
158 | if (!PageDirty(page)) | |
159 | pr_err("%p invalidatepage %p page not dirty\n", inode, page); | |
160 | ||
b150f5c1 | 161 | ClearPageChecked(page); |
1d3576fd | 162 | |
b150f5c1 MT |
163 | dout("%p invalidatepage %p idx %lu full dirty page\n", |
164 | inode, page, page->index); | |
165 | ||
166 | ceph_put_wrbuffer_cap_refs(ci, 1, snapc); | |
167 | ceph_put_snap_context(snapc); | |
168 | page->private = 0; | |
169 | ClearPagePrivate(page); | |
1d3576fd SW |
170 | } |
171 | ||
1d3576fd SW |
172 | static int ceph_releasepage(struct page *page, gfp_t g) |
173 | { | |
174 | struct inode *inode = page->mapping ? page->mapping->host : NULL; | |
175 | dout("%p releasepage %p idx %lu\n", inode, page, page->index); | |
176 | WARN_ON(PageDirty(page)); | |
99ccbd22 MT |
177 | |
178 | /* Can we release the page from the cache? */ | |
179 | if (!ceph_release_fscache_page(page, g)) | |
180 | return 0; | |
181 | ||
182 | return !PagePrivate(page); | |
1d3576fd SW |
183 | } |
184 | ||
185 | /* | |
186 | * read a single page, without unlocking it. | |
187 | */ | |
188 | static int readpage_nounlock(struct file *filp, struct page *page) | |
189 | { | |
496ad9aa | 190 | struct inode *inode = file_inode(filp); |
1d3576fd | 191 | struct ceph_inode_info *ci = ceph_inode(inode); |
99ccbd22 | 192 | struct ceph_osd_client *osdc = |
3d14c5d2 | 193 | &ceph_inode_to_client(inode)->client->osdc; |
1d3576fd | 194 | int err = 0; |
83701246 | 195 | u64 off = page_offset(page); |
1d3576fd SW |
196 | u64 len = PAGE_CACHE_SIZE; |
197 | ||
83701246 | 198 | if (off >= i_size_read(inode)) { |
fcc02d2a | 199 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); |
83701246 YZ |
200 | SetPageUptodate(page); |
201 | return 0; | |
202 | } | |
99ccbd22 | 203 | |
fcc02d2a YZ |
204 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
205 | /* | |
206 | * Uptodate inline data should have been added | |
207 | * into page cache while getting Fcr caps. | |
208 | */ | |
209 | if (off == 0) | |
210 | return -EINVAL; | |
211 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
212 | SetPageUptodate(page); | |
213 | return 0; | |
214 | } | |
83701246 YZ |
215 | |
216 | err = ceph_readpage_from_fscache(inode, page); | |
99ccbd22 MT |
217 | if (err == 0) |
218 | goto out; | |
219 | ||
1d3576fd SW |
220 | dout("readpage inode %p file %p page %p index %lu\n", |
221 | inode, filp, page, page->index); | |
222 | err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout, | |
83701246 | 223 | off, &len, |
1d3576fd | 224 | ci->i_truncate_seq, ci->i_truncate_size, |
b7495fc2 | 225 | &page, 1, 0); |
1d3576fd SW |
226 | if (err == -ENOENT) |
227 | err = 0; | |
228 | if (err < 0) { | |
229 | SetPageError(page); | |
18302805 | 230 | ceph_fscache_readpage_cancel(inode, page); |
1d3576fd | 231 | goto out; |
1d3576fd | 232 | } |
23cd573b ZZ |
233 | if (err < PAGE_CACHE_SIZE) |
234 | /* zero fill remainder of page */ | |
235 | zero_user_segment(page, err, PAGE_CACHE_SIZE); | |
236 | else | |
237 | flush_dcache_page(page); | |
1d3576fd | 238 | |
23cd573b ZZ |
239 | SetPageUptodate(page); |
240 | ceph_readpage_to_fscache(inode, page); | |
99ccbd22 | 241 | |
1d3576fd SW |
242 | out: |
243 | return err < 0 ? err : 0; | |
244 | } | |
245 | ||
246 | static int ceph_readpage(struct file *filp, struct page *page) | |
247 | { | |
248 | int r = readpage_nounlock(filp, page); | |
249 | unlock_page(page); | |
250 | return r; | |
251 | } | |
252 | ||
253 | /* | |
7c272194 | 254 | * Finish an async read(ahead) op. |
1d3576fd | 255 | */ |
7c272194 | 256 | static void finish_read(struct ceph_osd_request *req, struct ceph_msg *msg) |
1d3576fd | 257 | { |
7c272194 | 258 | struct inode *inode = req->r_inode; |
87060c10 | 259 | struct ceph_osd_data *osd_data; |
1b83bef2 SW |
260 | int rc = req->r_result; |
261 | int bytes = le32_to_cpu(msg->hdr.data_len); | |
e0c59487 | 262 | int num_pages; |
7c272194 | 263 | int i; |
1d3576fd | 264 | |
7c272194 SW |
265 | dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes); |
266 | ||
267 | /* unlock all pages, zeroing any data we didn't read */ | |
406e2c9f | 268 | osd_data = osd_req_op_extent_osd_data(req, 0); |
87060c10 AE |
269 | BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); |
270 | num_pages = calc_pages_for((u64)osd_data->alignment, | |
271 | (u64)osd_data->length); | |
e0c59487 | 272 | for (i = 0; i < num_pages; i++) { |
87060c10 | 273 | struct page *page = osd_data->pages[i]; |
7c272194 | 274 | |
f36132a7 LW |
275 | if (rc < 0) |
276 | goto unlock; | |
7c272194 SW |
277 | if (bytes < (int)PAGE_CACHE_SIZE) { |
278 | /* zero (remainder of) page */ | |
279 | int s = bytes < 0 ? 0 : bytes; | |
280 | zero_user_segment(page, s, PAGE_CACHE_SIZE); | |
1d3576fd | 281 | } |
7c272194 SW |
282 | dout("finish_read %p uptodate %p idx %lu\n", inode, page, |
283 | page->index); | |
284 | flush_dcache_page(page); | |
285 | SetPageUptodate(page); | |
99ccbd22 | 286 | ceph_readpage_to_fscache(inode, page); |
f36132a7 | 287 | unlock: |
7c272194 SW |
288 | unlock_page(page); |
289 | page_cache_release(page); | |
0fff87ec | 290 | bytes -= PAGE_CACHE_SIZE; |
1d3576fd | 291 | } |
87060c10 | 292 | kfree(osd_data->pages); |
1d3576fd SW |
293 | } |
294 | ||
8884d53d DZ |
295 | static void ceph_unlock_page_vector(struct page **pages, int num_pages) |
296 | { | |
297 | int i; | |
298 | ||
299 | for (i = 0; i < num_pages; i++) | |
300 | unlock_page(pages[i]); | |
301 | } | |
302 | ||
1d3576fd | 303 | /* |
7c272194 SW |
304 | * start an async read(ahead) operation. return nr_pages we submitted |
305 | * a read for on success, or negative error code. | |
1d3576fd | 306 | */ |
0d66a487 | 307 | static int start_read(struct inode *inode, struct list_head *page_list, int max) |
1d3576fd | 308 | { |
3d14c5d2 YS |
309 | struct ceph_osd_client *osdc = |
310 | &ceph_inode_to_client(inode)->client->osdc; | |
7c272194 SW |
311 | struct ceph_inode_info *ci = ceph_inode(inode); |
312 | struct page *page = list_entry(page_list->prev, struct page, lru); | |
acead002 | 313 | struct ceph_vino vino; |
7c272194 SW |
314 | struct ceph_osd_request *req; |
315 | u64 off; | |
1d3576fd | 316 | u64 len; |
7c272194 SW |
317 | int i; |
318 | struct page **pages; | |
319 | pgoff_t next_index; | |
320 | int nr_pages = 0; | |
321 | int ret; | |
1d3576fd | 322 | |
6285bc23 | 323 | off = (u64) page_offset(page); |
1d3576fd | 324 | |
7c272194 SW |
325 | /* count pages */ |
326 | next_index = page->index; | |
327 | list_for_each_entry_reverse(page, page_list, lru) { | |
328 | if (page->index != next_index) | |
329 | break; | |
330 | nr_pages++; | |
331 | next_index++; | |
0d66a487 SW |
332 | if (max && nr_pages == max) |
333 | break; | |
7c272194 | 334 | } |
1d3576fd | 335 | len = nr_pages << PAGE_CACHE_SHIFT; |
7c272194 SW |
336 | dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages, |
337 | off, len); | |
acead002 AE |
338 | vino = ceph_vino(inode); |
339 | req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len, | |
715e4cd4 | 340 | 0, 1, CEPH_OSD_OP_READ, |
acead002 | 341 | CEPH_OSD_FLAG_READ, NULL, |
7c272194 | 342 | ci->i_truncate_seq, ci->i_truncate_size, |
acead002 | 343 | false); |
6816282d SW |
344 | if (IS_ERR(req)) |
345 | return PTR_ERR(req); | |
1d3576fd | 346 | |
7c272194 | 347 | /* build page vector */ |
cf7b7e14 | 348 | nr_pages = calc_pages_for(0, len); |
7c272194 SW |
349 | pages = kmalloc(sizeof(*pages) * nr_pages, GFP_NOFS); |
350 | ret = -ENOMEM; | |
351 | if (!pages) | |
352 | goto out; | |
353 | for (i = 0; i < nr_pages; ++i) { | |
354 | page = list_entry(page_list->prev, struct page, lru); | |
355 | BUG_ON(PageLocked(page)); | |
1d3576fd | 356 | list_del(&page->lru); |
99ccbd22 | 357 | |
7c272194 SW |
358 | dout("start_read %p adding %p idx %lu\n", inode, page, |
359 | page->index); | |
360 | if (add_to_page_cache_lru(page, &inode->i_data, page->index, | |
213c99ee | 361 | GFP_NOFS)) { |
d4d3aa38 | 362 | ceph_fscache_uncache_page(inode, page); |
1d3576fd | 363 | page_cache_release(page); |
7c272194 | 364 | dout("start_read %p add_to_page_cache failed %p\n", |
1d3576fd | 365 | inode, page); |
7c272194 SW |
366 | nr_pages = i; |
367 | goto out_pages; | |
1d3576fd | 368 | } |
7c272194 | 369 | pages[i] = page; |
1d3576fd | 370 | } |
406e2c9f | 371 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false); |
7c272194 SW |
372 | req->r_callback = finish_read; |
373 | req->r_inode = inode; | |
374 | ||
79528734 | 375 | ceph_osdc_build_request(req, off, NULL, vino.snap, NULL); |
02ee07d3 | 376 | |
7c272194 SW |
377 | dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len); |
378 | ret = ceph_osdc_start_request(osdc, req, false); | |
379 | if (ret < 0) | |
380 | goto out_pages; | |
381 | ceph_osdc_put_request(req); | |
382 | return nr_pages; | |
383 | ||
384 | out_pages: | |
8884d53d | 385 | ceph_unlock_page_vector(pages, nr_pages); |
7c272194 | 386 | ceph_release_page_vector(pages, nr_pages); |
7c272194 SW |
387 | out: |
388 | ceph_osdc_put_request(req); | |
389 | return ret; | |
390 | } | |
1d3576fd | 391 | |
7c272194 SW |
392 | |
393 | /* | |
394 | * Read multiple pages. Leave pages we don't read + unlock in page_list; | |
395 | * the caller (VM) cleans them up. | |
396 | */ | |
397 | static int ceph_readpages(struct file *file, struct address_space *mapping, | |
398 | struct list_head *page_list, unsigned nr_pages) | |
399 | { | |
496ad9aa | 400 | struct inode *inode = file_inode(file); |
0d66a487 | 401 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
7c272194 | 402 | int rc = 0; |
0d66a487 SW |
403 | int max = 0; |
404 | ||
83701246 YZ |
405 | if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE) |
406 | return -EINVAL; | |
407 | ||
99ccbd22 MT |
408 | rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list, |
409 | &nr_pages); | |
410 | ||
411 | if (rc == 0) | |
412 | goto out; | |
413 | ||
0d66a487 SW |
414 | if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE) |
415 | max = (fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1) | |
416 | >> PAGE_SHIFT; | |
7c272194 | 417 | |
2794a82a AE |
418 | dout("readpages %p file %p nr_pages %d max %d\n", inode, |
419 | file, nr_pages, | |
0d66a487 | 420 | max); |
7c272194 | 421 | while (!list_empty(page_list)) { |
0d66a487 | 422 | rc = start_read(inode, page_list, max); |
7c272194 SW |
423 | if (rc < 0) |
424 | goto out; | |
425 | BUG_ON(rc == 0); | |
426 | } | |
1d3576fd | 427 | out: |
76be778b MT |
428 | ceph_fscache_readpages_cancel(inode, page_list); |
429 | ||
7c272194 | 430 | dout("readpages %p file %p ret %d\n", inode, file, rc); |
1d3576fd SW |
431 | return rc; |
432 | } | |
433 | ||
434 | /* | |
435 | * Get ref for the oldest snapc for an inode with dirty data... that is, the | |
436 | * only snap context we are allowed to write back. | |
1d3576fd | 437 | */ |
6298a337 SW |
438 | static struct ceph_snap_context *get_oldest_context(struct inode *inode, |
439 | u64 *snap_size) | |
1d3576fd SW |
440 | { |
441 | struct ceph_inode_info *ci = ceph_inode(inode); | |
442 | struct ceph_snap_context *snapc = NULL; | |
443 | struct ceph_cap_snap *capsnap = NULL; | |
444 | ||
be655596 | 445 | spin_lock(&ci->i_ceph_lock); |
1d3576fd SW |
446 | list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { |
447 | dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap, | |
448 | capsnap->context, capsnap->dirty_pages); | |
449 | if (capsnap->dirty_pages) { | |
450 | snapc = ceph_get_snap_context(capsnap->context); | |
451 | if (snap_size) | |
452 | *snap_size = capsnap->size; | |
453 | break; | |
454 | } | |
455 | } | |
7d8cb26d | 456 | if (!snapc && ci->i_wrbuffer_ref_head) { |
80e755fe | 457 | snapc = ceph_get_snap_context(ci->i_head_snapc); |
1d3576fd SW |
458 | dout(" head snapc %p has %d dirty pages\n", |
459 | snapc, ci->i_wrbuffer_ref_head); | |
460 | } | |
be655596 | 461 | spin_unlock(&ci->i_ceph_lock); |
1d3576fd SW |
462 | return snapc; |
463 | } | |
464 | ||
465 | /* | |
466 | * Write a single page, but leave the page locked. | |
467 | * | |
468 | * If we get a write error, set the page error bit, but still adjust the | |
469 | * dirty page accounting (i.e., page is no longer dirty). | |
470 | */ | |
471 | static int writepage_nounlock(struct page *page, struct writeback_control *wbc) | |
472 | { | |
473 | struct inode *inode; | |
474 | struct ceph_inode_info *ci; | |
3d14c5d2 | 475 | struct ceph_fs_client *fsc; |
1d3576fd | 476 | struct ceph_osd_client *osdc; |
6298a337 | 477 | struct ceph_snap_context *snapc, *oldest; |
fc2744aa | 478 | loff_t page_off = page_offset(page); |
2baba250 | 479 | long writeback_stat; |
fc2744aa YZ |
480 | u64 truncate_size, snap_size = 0; |
481 | u32 truncate_seq; | |
482 | int err = 0, len = PAGE_CACHE_SIZE; | |
1d3576fd SW |
483 | |
484 | dout("writepage %p idx %lu\n", page, page->index); | |
485 | ||
486 | if (!page->mapping || !page->mapping->host) { | |
487 | dout("writepage %p - no mapping\n", page); | |
488 | return -EFAULT; | |
489 | } | |
490 | inode = page->mapping->host; | |
491 | ci = ceph_inode(inode); | |
3d14c5d2 YS |
492 | fsc = ceph_inode_to_client(inode); |
493 | osdc = &fsc->client->osdc; | |
1d3576fd SW |
494 | |
495 | /* verify this is a writeable snap context */ | |
61600ef8 | 496 | snapc = page_snap_context(page); |
1d3576fd SW |
497 | if (snapc == NULL) { |
498 | dout("writepage %p page %p not dirty?\n", inode, page); | |
499 | goto out; | |
500 | } | |
6298a337 SW |
501 | oldest = get_oldest_context(inode, &snap_size); |
502 | if (snapc->seq > oldest->seq) { | |
1d3576fd | 503 | dout("writepage %p page %p snapc %p not writeable - noop\n", |
61600ef8 | 504 | inode, page, snapc); |
1d3576fd SW |
505 | /* we should only noop if called by kswapd */ |
506 | WARN_ON((current->flags & PF_MEMALLOC) == 0); | |
6298a337 | 507 | ceph_put_snap_context(oldest); |
1d3576fd SW |
508 | goto out; |
509 | } | |
6298a337 | 510 | ceph_put_snap_context(oldest); |
1d3576fd | 511 | |
fc2744aa YZ |
512 | spin_lock(&ci->i_ceph_lock); |
513 | truncate_seq = ci->i_truncate_seq; | |
514 | truncate_size = ci->i_truncate_size; | |
515 | if (!snap_size) | |
516 | snap_size = i_size_read(inode); | |
517 | spin_unlock(&ci->i_ceph_lock); | |
518 | ||
1d3576fd | 519 | /* is this a partial page at end of file? */ |
fc2744aa YZ |
520 | if (page_off >= snap_size) { |
521 | dout("%p page eof %llu\n", page, snap_size); | |
522 | goto out; | |
523 | } | |
524 | if (snap_size < page_off + len) | |
525 | len = snap_size - page_off; | |
1d3576fd | 526 | |
ae00d4f3 SW |
527 | dout("writepage %p page %p index %lu on %llu~%u snapc %p\n", |
528 | inode, page, page->index, page_off, len, snapc); | |
1d3576fd | 529 | |
3d14c5d2 | 530 | writeback_stat = atomic_long_inc_return(&fsc->writeback_count); |
2baba250 | 531 | if (writeback_stat > |
3d14c5d2 YS |
532 | CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb)) |
533 | set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC); | |
2baba250 | 534 | |
99ccbd22 MT |
535 | ceph_readpage_to_fscache(inode, page); |
536 | ||
1d3576fd SW |
537 | set_page_writeback(page); |
538 | err = ceph_osdc_writepages(osdc, ceph_vino(inode), | |
539 | &ci->i_layout, snapc, | |
540 | page_off, len, | |
fc2744aa | 541 | truncate_seq, truncate_size, |
24808826 | 542 | &inode->i_mtime, &page, 1); |
1d3576fd SW |
543 | if (err < 0) { |
544 | dout("writepage setting page/mapping error %d %p\n", err, page); | |
545 | SetPageError(page); | |
546 | mapping_set_error(&inode->i_data, err); | |
547 | if (wbc) | |
548 | wbc->pages_skipped++; | |
549 | } else { | |
550 | dout("writepage cleaned page %p\n", page); | |
551 | err = 0; /* vfs expects us to return 0 */ | |
552 | } | |
553 | page->private = 0; | |
554 | ClearPagePrivate(page); | |
555 | end_page_writeback(page); | |
556 | ceph_put_wrbuffer_cap_refs(ci, 1, snapc); | |
6298a337 | 557 | ceph_put_snap_context(snapc); /* page's reference */ |
1d3576fd SW |
558 | out: |
559 | return err; | |
560 | } | |
561 | ||
562 | static int ceph_writepage(struct page *page, struct writeback_control *wbc) | |
563 | { | |
dbd646a8 YS |
564 | int err; |
565 | struct inode *inode = page->mapping->host; | |
566 | BUG_ON(!inode); | |
70b666c3 | 567 | ihold(inode); |
dbd646a8 | 568 | err = writepage_nounlock(page, wbc); |
1d3576fd | 569 | unlock_page(page); |
dbd646a8 | 570 | iput(inode); |
1d3576fd SW |
571 | return err; |
572 | } | |
573 | ||
574 | ||
575 | /* | |
576 | * lame release_pages helper. release_pages() isn't exported to | |
577 | * modules. | |
578 | */ | |
579 | static void ceph_release_pages(struct page **pages, int num) | |
580 | { | |
581 | struct pagevec pvec; | |
582 | int i; | |
583 | ||
584 | pagevec_init(&pvec, 0); | |
585 | for (i = 0; i < num; i++) { | |
586 | if (pagevec_add(&pvec, pages[i]) == 0) | |
587 | pagevec_release(&pvec); | |
588 | } | |
589 | pagevec_release(&pvec); | |
590 | } | |
591 | ||
1d3576fd SW |
592 | /* |
593 | * async writeback completion handler. | |
594 | * | |
595 | * If we get an error, set the mapping error bit, but not the individual | |
596 | * page error bits. | |
597 | */ | |
598 | static void writepages_finish(struct ceph_osd_request *req, | |
599 | struct ceph_msg *msg) | |
600 | { | |
601 | struct inode *inode = req->r_inode; | |
1d3576fd | 602 | struct ceph_inode_info *ci = ceph_inode(inode); |
87060c10 | 603 | struct ceph_osd_data *osd_data; |
1d3576fd | 604 | unsigned wrote; |
1d3576fd | 605 | struct page *page; |
e0c59487 | 606 | int num_pages; |
1d3576fd SW |
607 | int i; |
608 | struct ceph_snap_context *snapc = req->r_snapc; | |
609 | struct address_space *mapping = inode->i_mapping; | |
1b83bef2 | 610 | int rc = req->r_result; |
79528734 | 611 | u64 bytes = req->r_ops[0].extent.length; |
3d14c5d2 | 612 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
2baba250 | 613 | long writeback_stat; |
7ff899da | 614 | unsigned issued = ceph_caps_issued(ci); |
1d3576fd | 615 | |
406e2c9f | 616 | osd_data = osd_req_op_extent_osd_data(req, 0); |
87060c10 AE |
617 | BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); |
618 | num_pages = calc_pages_for((u64)osd_data->alignment, | |
619 | (u64)osd_data->length); | |
1d3576fd | 620 | if (rc >= 0) { |
79788c69 SW |
621 | /* |
622 | * Assume we wrote the pages we originally sent. The | |
623 | * osd might reply with fewer pages if our writeback | |
624 | * raced with a truncation and was adjusted at the osd, | |
625 | * so don't believe the reply. | |
626 | */ | |
e0c59487 | 627 | wrote = num_pages; |
1d3576fd SW |
628 | } else { |
629 | wrote = 0; | |
630 | mapping_set_error(mapping, rc); | |
631 | } | |
632 | dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n", | |
633 | inode, rc, bytes, wrote); | |
634 | ||
635 | /* clean all pages */ | |
e0c59487 | 636 | for (i = 0; i < num_pages; i++) { |
87060c10 | 637 | page = osd_data->pages[i]; |
1d3576fd SW |
638 | BUG_ON(!page); |
639 | WARN_ON(!PageUptodate(page)); | |
640 | ||
2baba250 | 641 | writeback_stat = |
3d14c5d2 | 642 | atomic_long_dec_return(&fsc->writeback_count); |
2baba250 | 643 | if (writeback_stat < |
3d14c5d2 YS |
644 | CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb)) |
645 | clear_bdi_congested(&fsc->backing_dev_info, | |
2baba250 YS |
646 | BLK_RW_ASYNC); |
647 | ||
61600ef8 | 648 | ceph_put_snap_context(page_snap_context(page)); |
1d3576fd SW |
649 | page->private = 0; |
650 | ClearPagePrivate(page); | |
1d3576fd SW |
651 | dout("unlocking %d %p\n", i, page); |
652 | end_page_writeback(page); | |
e63dc5c7 YS |
653 | |
654 | /* | |
655 | * We lost the cache cap, need to truncate the page before | |
656 | * it is unlocked, otherwise we'd truncate it later in the | |
657 | * page truncation thread, possibly losing some data that | |
658 | * raced its way in | |
659 | */ | |
2962507c | 660 | if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) |
e63dc5c7 YS |
661 | generic_error_remove_page(inode->i_mapping, page); |
662 | ||
1d3576fd SW |
663 | unlock_page(page); |
664 | } | |
665 | dout("%p wrote+cleaned %d pages\n", inode, wrote); | |
e0c59487 | 666 | ceph_put_wrbuffer_cap_refs(ci, num_pages, snapc); |
1d3576fd | 667 | |
87060c10 AE |
668 | ceph_release_pages(osd_data->pages, num_pages); |
669 | if (osd_data->pages_from_pool) | |
670 | mempool_free(osd_data->pages, | |
640ef79d | 671 | ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool); |
1d3576fd | 672 | else |
87060c10 | 673 | kfree(osd_data->pages); |
1d3576fd SW |
674 | ceph_osdc_put_request(req); |
675 | } | |
676 | ||
1d3576fd SW |
677 | /* |
678 | * initiate async writeback | |
679 | */ | |
680 | static int ceph_writepages_start(struct address_space *mapping, | |
681 | struct writeback_control *wbc) | |
682 | { | |
683 | struct inode *inode = mapping->host; | |
1d3576fd | 684 | struct ceph_inode_info *ci = ceph_inode(inode); |
fc2744aa YZ |
685 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
686 | struct ceph_vino vino = ceph_vino(inode); | |
1d3576fd SW |
687 | pgoff_t index, start, end; |
688 | int range_whole = 0; | |
689 | int should_loop = 1; | |
690 | pgoff_t max_pages = 0, max_pages_ever = 0; | |
80e755fe | 691 | struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc; |
1d3576fd SW |
692 | struct pagevec pvec; |
693 | int done = 0; | |
694 | int rc = 0; | |
695 | unsigned wsize = 1 << inode->i_blkbits; | |
696 | struct ceph_osd_request *req = NULL; | |
021b77be | 697 | int do_sync = 0; |
fc2744aa YZ |
698 | u64 truncate_size, snap_size; |
699 | u32 truncate_seq; | |
1d3576fd SW |
700 | |
701 | /* | |
702 | * Include a 'sync' in the OSD request if this is a data | |
703 | * integrity write (e.g., O_SYNC write or fsync()), or if our | |
704 | * cap is being revoked. | |
705 | */ | |
c62988ec | 706 | if ((wbc->sync_mode == WB_SYNC_ALL) || |
707 | ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER)) | |
1d3576fd SW |
708 | do_sync = 1; |
709 | dout("writepages_start %p dosync=%d (mode=%s)\n", | |
710 | inode, do_sync, | |
711 | wbc->sync_mode == WB_SYNC_NONE ? "NONE" : | |
712 | (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD")); | |
713 | ||
3d14c5d2 | 714 | if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) { |
f3ae1b97 | 715 | pr_warn("writepage_start %p on forced umount\n", inode); |
1d3576fd SW |
716 | return -EIO; /* we're in a forced umount, don't write! */ |
717 | } | |
3d14c5d2 YS |
718 | if (fsc->mount_options->wsize && fsc->mount_options->wsize < wsize) |
719 | wsize = fsc->mount_options->wsize; | |
1d3576fd SW |
720 | if (wsize < PAGE_CACHE_SIZE) |
721 | wsize = PAGE_CACHE_SIZE; | |
722 | max_pages_ever = wsize >> PAGE_CACHE_SHIFT; | |
723 | ||
724 | pagevec_init(&pvec, 0); | |
725 | ||
1d3576fd SW |
726 | /* where to start/end? */ |
727 | if (wbc->range_cyclic) { | |
728 | start = mapping->writeback_index; /* Start from prev offset */ | |
729 | end = -1; | |
730 | dout(" cyclic, start at %lu\n", start); | |
731 | } else { | |
732 | start = wbc->range_start >> PAGE_CACHE_SHIFT; | |
733 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
734 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) | |
735 | range_whole = 1; | |
736 | should_loop = 0; | |
737 | dout(" not cyclic, %lu to %lu\n", start, end); | |
738 | } | |
739 | index = start; | |
740 | ||
741 | retry: | |
742 | /* find oldest snap context with dirty data */ | |
743 | ceph_put_snap_context(snapc); | |
1ac0fc8a | 744 | snap_size = 0; |
1d3576fd SW |
745 | snapc = get_oldest_context(inode, &snap_size); |
746 | if (!snapc) { | |
747 | /* hmm, why does writepages get called when there | |
748 | is no dirty data? */ | |
749 | dout(" no snap context with dirty data?\n"); | |
750 | goto out; | |
751 | } | |
1ac0fc8a YZ |
752 | if (snap_size == 0) |
753 | snap_size = i_size_read(inode); | |
1d3576fd SW |
754 | dout(" oldest snapc is %p seq %lld (%d snaps)\n", |
755 | snapc, snapc->seq, snapc->num_snaps); | |
fc2744aa YZ |
756 | |
757 | spin_lock(&ci->i_ceph_lock); | |
758 | truncate_seq = ci->i_truncate_seq; | |
759 | truncate_size = ci->i_truncate_size; | |
760 | if (!snap_size) | |
761 | snap_size = i_size_read(inode); | |
762 | spin_unlock(&ci->i_ceph_lock); | |
763 | ||
1d3576fd SW |
764 | if (last_snapc && snapc != last_snapc) { |
765 | /* if we switched to a newer snapc, restart our scan at the | |
766 | * start of the original file range. */ | |
767 | dout(" snapc differs from last pass, restarting at %lu\n", | |
768 | index); | |
769 | index = start; | |
770 | } | |
771 | last_snapc = snapc; | |
772 | ||
773 | while (!done && index <= end) { | |
774 | unsigned i; | |
775 | int first; | |
776 | pgoff_t next; | |
777 | int pvec_pages, locked_pages; | |
e5975c7c AE |
778 | struct page **pages = NULL; |
779 | mempool_t *pool = NULL; /* Becomes non-null if mempool used */ | |
1d3576fd SW |
780 | struct page *page; |
781 | int want; | |
782 | u64 offset, len; | |
2baba250 | 783 | long writeback_stat; |
1d3576fd SW |
784 | |
785 | next = 0; | |
786 | locked_pages = 0; | |
787 | max_pages = max_pages_ever; | |
788 | ||
789 | get_more_pages: | |
790 | first = -1; | |
791 | want = min(end - index, | |
792 | min((pgoff_t)PAGEVEC_SIZE, | |
793 | max_pages - (pgoff_t)locked_pages) - 1) | |
794 | + 1; | |
795 | pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
796 | PAGECACHE_TAG_DIRTY, | |
797 | want); | |
798 | dout("pagevec_lookup_tag got %d\n", pvec_pages); | |
799 | if (!pvec_pages && !locked_pages) | |
800 | break; | |
801 | for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) { | |
802 | page = pvec.pages[i]; | |
803 | dout("? %p idx %lu\n", page, page->index); | |
804 | if (locked_pages == 0) | |
805 | lock_page(page); /* first page */ | |
806 | else if (!trylock_page(page)) | |
807 | break; | |
808 | ||
809 | /* only dirty pages, or our accounting breaks */ | |
810 | if (unlikely(!PageDirty(page)) || | |
811 | unlikely(page->mapping != mapping)) { | |
812 | dout("!dirty or !mapping %p\n", page); | |
813 | unlock_page(page); | |
814 | break; | |
815 | } | |
816 | if (!wbc->range_cyclic && page->index > end) { | |
817 | dout("end of range %p\n", page); | |
818 | done = 1; | |
819 | unlock_page(page); | |
820 | break; | |
821 | } | |
822 | if (next && (page->index != next)) { | |
823 | dout("not consecutive %p\n", page); | |
824 | unlock_page(page); | |
825 | break; | |
826 | } | |
827 | if (wbc->sync_mode != WB_SYNC_NONE) { | |
828 | dout("waiting on writeback %p\n", page); | |
829 | wait_on_page_writeback(page); | |
830 | } | |
1ac0fc8a YZ |
831 | if (page_offset(page) >= snap_size) { |
832 | dout("%p page eof %llu\n", page, snap_size); | |
1d3576fd SW |
833 | done = 1; |
834 | unlock_page(page); | |
835 | break; | |
836 | } | |
837 | if (PageWriteback(page)) { | |
838 | dout("%p under writeback\n", page); | |
839 | unlock_page(page); | |
840 | break; | |
841 | } | |
842 | ||
843 | /* only if matching snap context */ | |
61600ef8 | 844 | pgsnapc = page_snap_context(page); |
80e755fe SW |
845 | if (pgsnapc->seq > snapc->seq) { |
846 | dout("page snapc %p %lld > oldest %p %lld\n", | |
847 | pgsnapc, pgsnapc->seq, snapc, snapc->seq); | |
1d3576fd SW |
848 | unlock_page(page); |
849 | if (!locked_pages) | |
850 | continue; /* keep looking for snap */ | |
851 | break; | |
852 | } | |
853 | ||
854 | if (!clear_page_dirty_for_io(page)) { | |
855 | dout("%p !clear_page_dirty_for_io\n", page); | |
856 | unlock_page(page); | |
857 | break; | |
858 | } | |
859 | ||
e5975c7c AE |
860 | /* |
861 | * We have something to write. If this is | |
862 | * the first locked page this time through, | |
863 | * allocate an osd request and a page array | |
864 | * that it will use. | |
865 | */ | |
1d3576fd | 866 | if (locked_pages == 0) { |
e5975c7c | 867 | BUG_ON(pages); |
1d3576fd | 868 | /* prepare async write request */ |
e5975c7c | 869 | offset = (u64)page_offset(page); |
1d3576fd | 870 | len = wsize; |
fc2744aa YZ |
871 | req = ceph_osdc_new_request(&fsc->client->osdc, |
872 | &ci->i_layout, vino, | |
715e4cd4 YZ |
873 | offset, &len, 0, |
874 | do_sync ? 2 : 1, | |
fc2744aa YZ |
875 | CEPH_OSD_OP_WRITE, |
876 | CEPH_OSD_FLAG_WRITE | | |
877 | CEPH_OSD_FLAG_ONDISK, | |
878 | snapc, truncate_seq, | |
879 | truncate_size, true); | |
6816282d SW |
880 | if (IS_ERR(req)) { |
881 | rc = PTR_ERR(req); | |
8c71897b HC |
882 | unlock_page(page); |
883 | break; | |
884 | } | |
885 | ||
715e4cd4 YZ |
886 | if (do_sync) |
887 | osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC); | |
888 | ||
88486957 AE |
889 | req->r_callback = writepages_finish; |
890 | req->r_inode = inode; | |
891 | ||
892 | max_pages = calc_pages_for(0, (u64)len); | |
fc2744aa YZ |
893 | pages = kmalloc(max_pages * sizeof (*pages), |
894 | GFP_NOFS); | |
88486957 AE |
895 | if (!pages) { |
896 | pool = fsc->wb_pagevec_pool; | |
88486957 | 897 | pages = mempool_alloc(pool, GFP_NOFS); |
e5975c7c | 898 | BUG_ON(!pages); |
88486957 | 899 | } |
1d3576fd SW |
900 | } |
901 | ||
902 | /* note position of first page in pvec */ | |
903 | if (first < 0) | |
904 | first = i; | |
905 | dout("%p will write page %p idx %lu\n", | |
906 | inode, page, page->index); | |
2baba250 | 907 | |
213c99ee | 908 | writeback_stat = |
3d14c5d2 | 909 | atomic_long_inc_return(&fsc->writeback_count); |
213c99ee | 910 | if (writeback_stat > CONGESTION_ON_THRESH( |
3d14c5d2 YS |
911 | fsc->mount_options->congestion_kb)) { |
912 | set_bdi_congested(&fsc->backing_dev_info, | |
213c99ee | 913 | BLK_RW_ASYNC); |
2baba250 YS |
914 | } |
915 | ||
1d3576fd | 916 | set_page_writeback(page); |
e5975c7c | 917 | pages[locked_pages] = page; |
1d3576fd SW |
918 | locked_pages++; |
919 | next = page->index + 1; | |
920 | } | |
921 | ||
922 | /* did we get anything? */ | |
923 | if (!locked_pages) | |
924 | goto release_pvec_pages; | |
925 | if (i) { | |
926 | int j; | |
927 | BUG_ON(!locked_pages || first < 0); | |
928 | ||
929 | if (pvec_pages && i == pvec_pages && | |
930 | locked_pages < max_pages) { | |
931 | dout("reached end pvec, trying for more\n"); | |
932 | pagevec_reinit(&pvec); | |
933 | goto get_more_pages; | |
934 | } | |
935 | ||
936 | /* shift unused pages over in the pvec... we | |
937 | * will need to release them below. */ | |
938 | for (j = i; j < pvec_pages; j++) { | |
939 | dout(" pvec leftover page %p\n", | |
940 | pvec.pages[j]); | |
941 | pvec.pages[j-i+first] = pvec.pages[j]; | |
942 | } | |
943 | pvec.nr -= i-first; | |
944 | } | |
945 | ||
e5975c7c AE |
946 | /* Format the osd request message and submit the write */ |
947 | ||
948 | offset = page_offset(pages[0]); | |
1ac0fc8a | 949 | len = min(snap_size - offset, |
1d3576fd SW |
950 | (u64)locked_pages << PAGE_CACHE_SHIFT); |
951 | dout("writepages got %d pages at %llu~%llu\n", | |
952 | locked_pages, offset, len); | |
953 | ||
406e2c9f | 954 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, |
a4ce40a9 | 955 | !!pool, false); |
e5975c7c AE |
956 | |
957 | pages = NULL; /* request message now owns the pages array */ | |
958 | pool = NULL; | |
959 | ||
960 | /* Update the write op length in case we changed it */ | |
961 | ||
c99d2d4a | 962 | osd_req_op_extent_update(req, 0, len); |
e5975c7c AE |
963 | |
964 | vino = ceph_vino(inode); | |
79528734 AE |
965 | ceph_osdc_build_request(req, offset, snapc, vino.snap, |
966 | &inode->i_mtime); | |
1d3576fd | 967 | |
9d6fcb08 SW |
968 | rc = ceph_osdc_start_request(&fsc->client->osdc, req, true); |
969 | BUG_ON(rc); | |
1d3576fd SW |
970 | req = NULL; |
971 | ||
972 | /* continue? */ | |
973 | index = next; | |
974 | wbc->nr_to_write -= locked_pages; | |
975 | if (wbc->nr_to_write <= 0) | |
976 | done = 1; | |
977 | ||
978 | release_pvec_pages: | |
979 | dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr, | |
980 | pvec.nr ? pvec.pages[0] : NULL); | |
981 | pagevec_release(&pvec); | |
982 | ||
983 | if (locked_pages && !done) | |
984 | goto retry; | |
985 | } | |
986 | ||
987 | if (should_loop && !done) { | |
988 | /* more to do; loop back to beginning of file */ | |
989 | dout("writepages looping back to beginning of file\n"); | |
990 | should_loop = 0; | |
991 | index = 0; | |
992 | goto retry; | |
993 | } | |
994 | ||
995 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) | |
996 | mapping->writeback_index = index; | |
997 | ||
998 | out: | |
999 | if (req) | |
1000 | ceph_osdc_put_request(req); | |
1d3576fd SW |
1001 | ceph_put_snap_context(snapc); |
1002 | dout("writepages done, rc = %d\n", rc); | |
1d3576fd SW |
1003 | return rc; |
1004 | } | |
1005 | ||
1006 | ||
1007 | ||
1008 | /* | |
1009 | * See if a given @snapc is either writeable, or already written. | |
1010 | */ | |
1011 | static int context_is_writeable_or_written(struct inode *inode, | |
1012 | struct ceph_snap_context *snapc) | |
1013 | { | |
1014 | struct ceph_snap_context *oldest = get_oldest_context(inode, NULL); | |
6298a337 SW |
1015 | int ret = !oldest || snapc->seq <= oldest->seq; |
1016 | ||
1017 | ceph_put_snap_context(oldest); | |
1018 | return ret; | |
1d3576fd SW |
1019 | } |
1020 | ||
1021 | /* | |
1022 | * We are only allowed to write into/dirty the page if the page is | |
1023 | * clean, or already dirty within the same snap context. | |
8f883c24 SW |
1024 | * |
1025 | * called with page locked. | |
1026 | * return success with page locked, | |
1027 | * or any failure (incl -EAGAIN) with page unlocked. | |
1d3576fd | 1028 | */ |
4af6b225 YS |
1029 | static int ceph_update_writeable_page(struct file *file, |
1030 | loff_t pos, unsigned len, | |
1031 | struct page *page) | |
1d3576fd | 1032 | { |
496ad9aa | 1033 | struct inode *inode = file_inode(file); |
1d3576fd | 1034 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 | 1035 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
1d3576fd SW |
1036 | loff_t page_off = pos & PAGE_CACHE_MASK; |
1037 | int pos_in_page = pos & ~PAGE_CACHE_MASK; | |
1038 | int end_in_page = pos_in_page + len; | |
1039 | loff_t i_size; | |
1d3576fd | 1040 | int r; |
80e755fe | 1041 | struct ceph_snap_context *snapc, *oldest; |
1d3576fd | 1042 | |
1d3576fd SW |
1043 | retry_locked: |
1044 | /* writepages currently holds page lock, but if we change that later, */ | |
1045 | wait_on_page_writeback(page); | |
1046 | ||
1047 | /* check snap context */ | |
1048 | BUG_ON(!ci->i_snap_realm); | |
1049 | down_read(&mdsc->snap_rwsem); | |
1050 | BUG_ON(!ci->i_snap_realm->cached_context); | |
61600ef8 | 1051 | snapc = page_snap_context(page); |
80e755fe | 1052 | if (snapc && snapc != ci->i_head_snapc) { |
1d3576fd SW |
1053 | /* |
1054 | * this page is already dirty in another (older) snap | |
1055 | * context! is it writeable now? | |
1056 | */ | |
80e755fe | 1057 | oldest = get_oldest_context(inode, NULL); |
1d3576fd SW |
1058 | up_read(&mdsc->snap_rwsem); |
1059 | ||
80e755fe | 1060 | if (snapc->seq > oldest->seq) { |
6298a337 | 1061 | ceph_put_snap_context(oldest); |
1d3576fd | 1062 | dout(" page %p snapc %p not current or oldest\n", |
6298a337 | 1063 | page, snapc); |
1d3576fd SW |
1064 | /* |
1065 | * queue for writeback, and wait for snapc to | |
1066 | * be writeable or written | |
1067 | */ | |
6298a337 | 1068 | snapc = ceph_get_snap_context(snapc); |
1d3576fd | 1069 | unlock_page(page); |
3c6f6b79 | 1070 | ceph_queue_writeback(inode); |
8f883c24 | 1071 | r = wait_event_interruptible(ci->i_cap_wq, |
1d3576fd SW |
1072 | context_is_writeable_or_written(inode, snapc)); |
1073 | ceph_put_snap_context(snapc); | |
8f883c24 SW |
1074 | if (r == -ERESTARTSYS) |
1075 | return r; | |
4af6b225 | 1076 | return -EAGAIN; |
1d3576fd | 1077 | } |
6298a337 | 1078 | ceph_put_snap_context(oldest); |
1d3576fd SW |
1079 | |
1080 | /* yay, writeable, do it now (without dropping page lock) */ | |
1081 | dout(" page %p snapc %p not current, but oldest\n", | |
1082 | page, snapc); | |
1083 | if (!clear_page_dirty_for_io(page)) | |
1084 | goto retry_locked; | |
1085 | r = writepage_nounlock(page, NULL); | |
1086 | if (r < 0) | |
1087 | goto fail_nosnap; | |
1088 | goto retry_locked; | |
1089 | } | |
1090 | ||
1091 | if (PageUptodate(page)) { | |
1092 | dout(" page %p already uptodate\n", page); | |
1093 | return 0; | |
1094 | } | |
1095 | ||
1096 | /* full page? */ | |
1097 | if (pos_in_page == 0 && len == PAGE_CACHE_SIZE) | |
1098 | return 0; | |
1099 | ||
1100 | /* past end of file? */ | |
1101 | i_size = inode->i_size; /* caller holds i_mutex */ | |
1102 | ||
1d3576fd SW |
1103 | if (page_off >= i_size || |
1104 | (pos_in_page == 0 && (pos+len) >= i_size && | |
1105 | end_in_page - pos_in_page != PAGE_CACHE_SIZE)) { | |
1106 | dout(" zeroing %p 0 - %d and %d - %d\n", | |
1107 | page, pos_in_page, end_in_page, (int)PAGE_CACHE_SIZE); | |
1108 | zero_user_segments(page, | |
1109 | 0, pos_in_page, | |
1110 | end_in_page, PAGE_CACHE_SIZE); | |
1111 | return 0; | |
1112 | } | |
1113 | ||
1114 | /* we need to read it. */ | |
1115 | up_read(&mdsc->snap_rwsem); | |
1116 | r = readpage_nounlock(file, page); | |
1117 | if (r < 0) | |
1118 | goto fail_nosnap; | |
1119 | goto retry_locked; | |
1d3576fd SW |
1120 | fail_nosnap: |
1121 | unlock_page(page); | |
1122 | return r; | |
1123 | } | |
1124 | ||
4af6b225 YS |
1125 | /* |
1126 | * We are only allowed to write into/dirty the page if the page is | |
1127 | * clean, or already dirty within the same snap context. | |
1128 | */ | |
1129 | static int ceph_write_begin(struct file *file, struct address_space *mapping, | |
1130 | loff_t pos, unsigned len, unsigned flags, | |
1131 | struct page **pagep, void **fsdata) | |
1132 | { | |
496ad9aa | 1133 | struct inode *inode = file_inode(file); |
4af6b225 YS |
1134 | struct page *page; |
1135 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
7971bd92 | 1136 | int r; |
4af6b225 YS |
1137 | |
1138 | do { | |
8f883c24 | 1139 | /* get a page */ |
4af6b225 | 1140 | page = grab_cache_page_write_begin(mapping, index, 0); |
7971bd92 SW |
1141 | if (!page) |
1142 | return -ENOMEM; | |
1143 | *pagep = page; | |
4af6b225 YS |
1144 | |
1145 | dout("write_begin file %p inode %p page %p %d~%d\n", file, | |
213c99ee | 1146 | inode, page, (int)pos, (int)len); |
4af6b225 YS |
1147 | |
1148 | r = ceph_update_writeable_page(file, pos, len, page); | |
1149 | } while (r == -EAGAIN); | |
1150 | ||
1151 | return r; | |
1152 | } | |
1153 | ||
1d3576fd SW |
1154 | /* |
1155 | * we don't do anything in here that simple_write_end doesn't do | |
1156 | * except adjust dirty page accounting and drop read lock on | |
1157 | * mdsc->snap_rwsem. | |
1158 | */ | |
1159 | static int ceph_write_end(struct file *file, struct address_space *mapping, | |
1160 | loff_t pos, unsigned len, unsigned copied, | |
1161 | struct page *page, void *fsdata) | |
1162 | { | |
496ad9aa | 1163 | struct inode *inode = file_inode(file); |
3d14c5d2 YS |
1164 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
1165 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
1d3576fd SW |
1166 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); |
1167 | int check_cap = 0; | |
1168 | ||
1169 | dout("write_end file %p inode %p page %p %d~%d (%d)\n", file, | |
1170 | inode, page, (int)pos, (int)copied, (int)len); | |
1171 | ||
1172 | /* zero the stale part of the page if we did a short copy */ | |
1173 | if (copied < len) | |
1174 | zero_user_segment(page, from+copied, len); | |
1175 | ||
1176 | /* did file size increase? */ | |
1177 | /* (no need for i_size_read(); we caller holds i_mutex */ | |
1178 | if (pos+copied > inode->i_size) | |
1179 | check_cap = ceph_inode_set_size(inode, pos+copied); | |
1180 | ||
1181 | if (!PageUptodate(page)) | |
1182 | SetPageUptodate(page); | |
1183 | ||
1184 | set_page_dirty(page); | |
1185 | ||
1186 | unlock_page(page); | |
1187 | up_read(&mdsc->snap_rwsem); | |
1188 | page_cache_release(page); | |
1189 | ||
1190 | if (check_cap) | |
1191 | ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL); | |
1192 | ||
1193 | return copied; | |
1194 | } | |
1195 | ||
1196 | /* | |
1197 | * we set .direct_IO to indicate direct io is supported, but since we | |
1198 | * intercept O_DIRECT reads and writes early, this function should | |
1199 | * never get called. | |
1200 | */ | |
1201 | static ssize_t ceph_direct_io(int rw, struct kiocb *iocb, | |
d8d3d94b AV |
1202 | struct iov_iter *iter, |
1203 | loff_t pos) | |
1d3576fd SW |
1204 | { |
1205 | WARN_ON(1); | |
1206 | return -EINVAL; | |
1207 | } | |
1208 | ||
1209 | const struct address_space_operations ceph_aops = { | |
1210 | .readpage = ceph_readpage, | |
1211 | .readpages = ceph_readpages, | |
1212 | .writepage = ceph_writepage, | |
1213 | .writepages = ceph_writepages_start, | |
1214 | .write_begin = ceph_write_begin, | |
1215 | .write_end = ceph_write_end, | |
1216 | .set_page_dirty = ceph_set_page_dirty, | |
1217 | .invalidatepage = ceph_invalidatepage, | |
1218 | .releasepage = ceph_releasepage, | |
1219 | .direct_IO = ceph_direct_io, | |
1220 | }; | |
1221 | ||
1222 | ||
1223 | /* | |
1224 | * vm ops | |
1225 | */ | |
61f68816 YZ |
1226 | static int ceph_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1227 | { | |
1228 | struct inode *inode = file_inode(vma->vm_file); | |
1229 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1230 | struct ceph_file_info *fi = vma->vm_file->private_data; | |
3738daa6 | 1231 | struct page *pinned_page = NULL; |
61f68816 YZ |
1232 | loff_t off = vmf->pgoff << PAGE_CACHE_SHIFT; |
1233 | int want, got, ret; | |
1234 | ||
1235 | dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n", | |
37b52fe6 | 1236 | inode, ceph_vinop(inode), off, (size_t)PAGE_CACHE_SIZE); |
61f68816 YZ |
1237 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
1238 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
1239 | else | |
1240 | want = CEPH_CAP_FILE_CACHE; | |
1241 | while (1) { | |
1242 | got = 0; | |
83701246 YZ |
1243 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, |
1244 | -1, &got, &pinned_page); | |
61f68816 YZ |
1245 | if (ret == 0) |
1246 | break; | |
1247 | if (ret != -ERESTARTSYS) { | |
1248 | WARN_ON(1); | |
1249 | return VM_FAULT_SIGBUS; | |
1250 | } | |
1251 | } | |
1252 | dout("filemap_fault %p %llu~%zd got cap refs on %s\n", | |
37b52fe6 | 1253 | inode, off, (size_t)PAGE_CACHE_SIZE, ceph_cap_string(got)); |
61f68816 | 1254 | |
83701246 YZ |
1255 | if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) || |
1256 | ci->i_inline_version == CEPH_INLINE_NONE) | |
1257 | ret = filemap_fault(vma, vmf); | |
1258 | else | |
1259 | ret = -EAGAIN; | |
61f68816 YZ |
1260 | |
1261 | dout("filemap_fault %p %llu~%zd dropping cap refs on %s ret %d\n", | |
37b52fe6 | 1262 | inode, off, (size_t)PAGE_CACHE_SIZE, ceph_cap_string(got), ret); |
3738daa6 YZ |
1263 | if (pinned_page) |
1264 | page_cache_release(pinned_page); | |
61f68816 YZ |
1265 | ceph_put_cap_refs(ci, got); |
1266 | ||
83701246 YZ |
1267 | if (ret != -EAGAIN) |
1268 | return ret; | |
1269 | ||
1270 | /* read inline data */ | |
1271 | if (off >= PAGE_CACHE_SIZE) { | |
1272 | /* does not support inline data > PAGE_SIZE */ | |
1273 | ret = VM_FAULT_SIGBUS; | |
1274 | } else { | |
1275 | int ret1; | |
1276 | struct address_space *mapping = inode->i_mapping; | |
1277 | struct page *page = find_or_create_page(mapping, 0, | |
1278 | mapping_gfp_mask(mapping) & | |
1279 | ~__GFP_FS); | |
1280 | if (!page) { | |
1281 | ret = VM_FAULT_OOM; | |
1282 | goto out; | |
1283 | } | |
1284 | ret1 = __ceph_do_getattr(inode, page, | |
1285 | CEPH_STAT_CAP_INLINE_DATA, true); | |
1286 | if (ret1 < 0 || off >= i_size_read(inode)) { | |
1287 | unlock_page(page); | |
1288 | page_cache_release(page); | |
1289 | ret = VM_FAULT_SIGBUS; | |
1290 | goto out; | |
1291 | } | |
1292 | if (ret1 < PAGE_CACHE_SIZE) | |
1293 | zero_user_segment(page, ret1, PAGE_CACHE_SIZE); | |
1294 | else | |
1295 | flush_dcache_page(page); | |
1296 | SetPageUptodate(page); | |
1297 | vmf->page = page; | |
1298 | ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED; | |
1299 | } | |
1300 | out: | |
1301 | dout("filemap_fault %p %llu~%zd read inline data ret %d\n", | |
1302 | inode, off, (size_t)PAGE_CACHE_SIZE, ret); | |
61f68816 YZ |
1303 | return ret; |
1304 | } | |
1d3576fd SW |
1305 | |
1306 | /* | |
1307 | * Reuse write_begin here for simplicity. | |
1308 | */ | |
1309 | static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) | |
1310 | { | |
496ad9aa | 1311 | struct inode *inode = file_inode(vma->vm_file); |
61f68816 YZ |
1312 | struct ceph_inode_info *ci = ceph_inode(inode); |
1313 | struct ceph_file_info *fi = vma->vm_file->private_data; | |
3d14c5d2 | 1314 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
61f68816 | 1315 | struct page *page = vmf->page; |
6285bc23 | 1316 | loff_t off = page_offset(page); |
61f68816 YZ |
1317 | loff_t size = i_size_read(inode); |
1318 | size_t len; | |
1319 | int want, got, ret; | |
3ca9c3bd | 1320 | |
28127bdd YZ |
1321 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
1322 | struct page *locked_page = NULL; | |
1323 | if (off == 0) { | |
1324 | lock_page(page); | |
1325 | locked_page = page; | |
1326 | } | |
1327 | ret = ceph_uninline_data(vma->vm_file, locked_page); | |
1328 | if (locked_page) | |
1329 | unlock_page(locked_page); | |
1330 | if (ret < 0) | |
1331 | return VM_FAULT_SIGBUS; | |
1332 | } | |
1333 | ||
1d3576fd SW |
1334 | if (off + PAGE_CACHE_SIZE <= size) |
1335 | len = PAGE_CACHE_SIZE; | |
1336 | else | |
1337 | len = size & ~PAGE_CACHE_MASK; | |
1338 | ||
61f68816 YZ |
1339 | dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n", |
1340 | inode, ceph_vinop(inode), off, len, size); | |
1341 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1342 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1343 | else | |
1344 | want = CEPH_CAP_FILE_BUFFER; | |
1345 | while (1) { | |
1346 | got = 0; | |
3738daa6 YZ |
1347 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, off + len, |
1348 | &got, NULL); | |
61f68816 YZ |
1349 | if (ret == 0) |
1350 | break; | |
1351 | if (ret != -ERESTARTSYS) { | |
1352 | WARN_ON(1); | |
1353 | return VM_FAULT_SIGBUS; | |
1354 | } | |
1355 | } | |
1356 | dout("page_mkwrite %p %llu~%zd got cap refs on %s\n", | |
1357 | inode, off, len, ceph_cap_string(got)); | |
1358 | ||
1359 | /* Update time before taking page lock */ | |
1360 | file_update_time(vma->vm_file); | |
4af6b225 YS |
1361 | |
1362 | lock_page(page); | |
1363 | ||
1364 | ret = VM_FAULT_NOPAGE; | |
1365 | if ((off > size) || | |
1366 | (page->mapping != inode->i_mapping)) | |
1367 | goto out; | |
1368 | ||
1369 | ret = ceph_update_writeable_page(vma->vm_file, off, len, page); | |
1370 | if (ret == 0) { | |
1371 | /* success. we'll keep the page locked. */ | |
1d3576fd SW |
1372 | set_page_dirty(page); |
1373 | up_read(&mdsc->snap_rwsem); | |
1d3576fd SW |
1374 | ret = VM_FAULT_LOCKED; |
1375 | } else { | |
4af6b225 YS |
1376 | if (ret == -ENOMEM) |
1377 | ret = VM_FAULT_OOM; | |
1378 | else | |
1379 | ret = VM_FAULT_SIGBUS; | |
1d3576fd | 1380 | } |
4af6b225 | 1381 | out: |
28127bdd | 1382 | if (ret != VM_FAULT_LOCKED) |
4af6b225 | 1383 | unlock_page(page); |
28127bdd YZ |
1384 | if (ret == VM_FAULT_LOCKED || |
1385 | ci->i_inline_version != CEPH_INLINE_NONE) { | |
61f68816 YZ |
1386 | int dirty; |
1387 | spin_lock(&ci->i_ceph_lock); | |
28127bdd | 1388 | ci->i_inline_version = CEPH_INLINE_NONE; |
61f68816 YZ |
1389 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR); |
1390 | spin_unlock(&ci->i_ceph_lock); | |
1391 | if (dirty) | |
1392 | __mark_inode_dirty(inode, dirty); | |
1393 | } | |
1394 | ||
1395 | dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %d\n", | |
1396 | inode, off, len, ceph_cap_string(got), ret); | |
1397 | ceph_put_cap_refs(ci, got); | |
1398 | ||
1d3576fd SW |
1399 | return ret; |
1400 | } | |
1401 | ||
31c542a1 YZ |
1402 | void ceph_fill_inline_data(struct inode *inode, struct page *locked_page, |
1403 | char *data, size_t len) | |
1404 | { | |
1405 | struct address_space *mapping = inode->i_mapping; | |
1406 | struct page *page; | |
1407 | ||
1408 | if (locked_page) { | |
1409 | page = locked_page; | |
1410 | } else { | |
1411 | if (i_size_read(inode) == 0) | |
1412 | return; | |
1413 | page = find_or_create_page(mapping, 0, | |
1414 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
1415 | if (!page) | |
1416 | return; | |
1417 | if (PageUptodate(page)) { | |
1418 | unlock_page(page); | |
1419 | page_cache_release(page); | |
1420 | return; | |
1421 | } | |
1422 | } | |
1423 | ||
0668ff52 | 1424 | dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n", |
31c542a1 YZ |
1425 | inode, ceph_vinop(inode), len, locked_page); |
1426 | ||
1427 | if (len > 0) { | |
1428 | void *kaddr = kmap_atomic(page); | |
1429 | memcpy(kaddr, data, len); | |
1430 | kunmap_atomic(kaddr); | |
1431 | } | |
1432 | ||
1433 | if (page != locked_page) { | |
1434 | if (len < PAGE_CACHE_SIZE) | |
1435 | zero_user_segment(page, len, PAGE_CACHE_SIZE); | |
1436 | else | |
1437 | flush_dcache_page(page); | |
1438 | ||
1439 | SetPageUptodate(page); | |
1440 | unlock_page(page); | |
1441 | page_cache_release(page); | |
1442 | } | |
1443 | } | |
1444 | ||
28127bdd YZ |
1445 | int ceph_uninline_data(struct file *filp, struct page *locked_page) |
1446 | { | |
1447 | struct inode *inode = file_inode(filp); | |
1448 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1449 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
1450 | struct ceph_osd_request *req; | |
1451 | struct page *page = NULL; | |
1452 | u64 len, inline_version; | |
1453 | int err = 0; | |
1454 | bool from_pagecache = false; | |
1455 | ||
1456 | spin_lock(&ci->i_ceph_lock); | |
1457 | inline_version = ci->i_inline_version; | |
1458 | spin_unlock(&ci->i_ceph_lock); | |
1459 | ||
1460 | dout("uninline_data %p %llx.%llx inline_version %llu\n", | |
1461 | inode, ceph_vinop(inode), inline_version); | |
1462 | ||
1463 | if (inline_version == 1 || /* initial version, no data */ | |
1464 | inline_version == CEPH_INLINE_NONE) | |
1465 | goto out; | |
1466 | ||
1467 | if (locked_page) { | |
1468 | page = locked_page; | |
1469 | WARN_ON(!PageUptodate(page)); | |
1470 | } else if (ceph_caps_issued(ci) & | |
1471 | (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) { | |
1472 | page = find_get_page(inode->i_mapping, 0); | |
1473 | if (page) { | |
1474 | if (PageUptodate(page)) { | |
1475 | from_pagecache = true; | |
1476 | lock_page(page); | |
1477 | } else { | |
1478 | page_cache_release(page); | |
1479 | page = NULL; | |
1480 | } | |
1481 | } | |
1482 | } | |
1483 | ||
1484 | if (page) { | |
1485 | len = i_size_read(inode); | |
1486 | if (len > PAGE_CACHE_SIZE) | |
1487 | len = PAGE_CACHE_SIZE; | |
1488 | } else { | |
1489 | page = __page_cache_alloc(GFP_NOFS); | |
1490 | if (!page) { | |
1491 | err = -ENOMEM; | |
1492 | goto out; | |
1493 | } | |
1494 | err = __ceph_do_getattr(inode, page, | |
1495 | CEPH_STAT_CAP_INLINE_DATA, true); | |
1496 | if (err < 0) { | |
1497 | /* no inline data */ | |
1498 | if (err == -ENODATA) | |
1499 | err = 0; | |
1500 | goto out; | |
1501 | } | |
1502 | len = err; | |
1503 | } | |
1504 | ||
1505 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1506 | ceph_vino(inode), 0, &len, 0, 1, | |
1507 | CEPH_OSD_OP_CREATE, | |
1508 | CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE, | |
1509 | ci->i_snap_realm->cached_context, | |
1510 | 0, 0, false); | |
1511 | if (IS_ERR(req)) { | |
1512 | err = PTR_ERR(req); | |
1513 | goto out; | |
1514 | } | |
1515 | ||
1516 | ceph_osdc_build_request(req, 0, NULL, CEPH_NOSNAP, &inode->i_mtime); | |
1517 | err = ceph_osdc_start_request(&fsc->client->osdc, req, false); | |
1518 | if (!err) | |
1519 | err = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1520 | ceph_osdc_put_request(req); | |
1521 | if (err < 0) | |
1522 | goto out; | |
1523 | ||
1524 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1525 | ceph_vino(inode), 0, &len, 1, 3, | |
1526 | CEPH_OSD_OP_WRITE, | |
1527 | CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE, | |
1528 | ci->i_snap_realm->cached_context, | |
1529 | ci->i_truncate_seq, ci->i_truncate_size, | |
1530 | false); | |
1531 | if (IS_ERR(req)) { | |
1532 | err = PTR_ERR(req); | |
1533 | goto out; | |
1534 | } | |
1535 | ||
1536 | osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false); | |
1537 | ||
1538 | err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR, | |
1539 | "inline_version", &inline_version, | |
1540 | sizeof(inline_version), | |
1541 | CEPH_OSD_CMPXATTR_OP_GT, | |
1542 | CEPH_OSD_CMPXATTR_MODE_U64); | |
1543 | if (err) | |
1544 | goto out_put; | |
1545 | ||
1546 | err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR, | |
1547 | "inline_version", &inline_version, | |
1548 | sizeof(inline_version), 0, 0); | |
1549 | if (err) | |
1550 | goto out_put; | |
1551 | ||
1552 | ceph_osdc_build_request(req, 0, NULL, CEPH_NOSNAP, &inode->i_mtime); | |
1553 | err = ceph_osdc_start_request(&fsc->client->osdc, req, false); | |
1554 | if (!err) | |
1555 | err = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1556 | out_put: | |
1557 | ceph_osdc_put_request(req); | |
1558 | if (err == -ECANCELED) | |
1559 | err = 0; | |
1560 | out: | |
1561 | if (page && page != locked_page) { | |
1562 | if (from_pagecache) { | |
1563 | unlock_page(page); | |
1564 | page_cache_release(page); | |
1565 | } else | |
1566 | __free_pages(page, 0); | |
1567 | } | |
1568 | ||
1569 | dout("uninline_data %p %llx.%llx inline_version %llu = %d\n", | |
1570 | inode, ceph_vinop(inode), inline_version, err); | |
1571 | return err; | |
1572 | } | |
1573 | ||
1d3576fd | 1574 | static struct vm_operations_struct ceph_vmops = { |
61f68816 | 1575 | .fault = ceph_filemap_fault, |
1d3576fd SW |
1576 | .page_mkwrite = ceph_page_mkwrite, |
1577 | }; | |
1578 | ||
1579 | int ceph_mmap(struct file *file, struct vm_area_struct *vma) | |
1580 | { | |
1581 | struct address_space *mapping = file->f_mapping; | |
1582 | ||
1583 | if (!mapping->a_ops->readpage) | |
1584 | return -ENOEXEC; | |
1585 | file_accessed(file); | |
1586 | vma->vm_ops = &ceph_vmops; | |
1d3576fd SW |
1587 | return 0; |
1588 | } |