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> | |
f361bf4a | 11 | #include <linux/signal.h> |
1d3576fd SW |
12 | |
13 | #include "super.h" | |
3d14c5d2 | 14 | #include "mds_client.h" |
99ccbd22 | 15 | #include "cache.h" |
3d14c5d2 | 16 | #include <linux/ceph/osd_client.h> |
1d3576fd SW |
17 | |
18 | /* | |
19 | * Ceph address space ops. | |
20 | * | |
21 | * There are a few funny things going on here. | |
22 | * | |
23 | * The page->private field is used to reference a struct | |
24 | * ceph_snap_context for _every_ dirty page. This indicates which | |
25 | * snapshot the page was logically dirtied in, and thus which snap | |
26 | * context needs to be associated with the osd write during writeback. | |
27 | * | |
28 | * Similarly, struct ceph_inode_info maintains a set of counters to | |
25985edc | 29 | * count dirty pages on the inode. In the absence of snapshots, |
1d3576fd SW |
30 | * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count. |
31 | * | |
32 | * When a snapshot is taken (that is, when the client receives | |
33 | * notification that a snapshot was taken), each inode with caps and | |
34 | * with dirty pages (dirty pages implies there is a cap) gets a new | |
35 | * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending | |
36 | * order, new snaps go to the tail). The i_wrbuffer_ref_head count is | |
37 | * moved to capsnap->dirty. (Unless a sync write is currently in | |
38 | * progress. In that case, the capsnap is said to be "pending", new | |
39 | * writes cannot start, and the capsnap isn't "finalized" until the | |
40 | * write completes (or fails) and a final size/mtime for the inode for | |
41 | * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0. | |
42 | * | |
43 | * On writeback, we must submit writes to the osd IN SNAP ORDER. So, | |
44 | * we look for the first capsnap in i_cap_snaps and write out pages in | |
45 | * that snap context _only_. Then we move on to the next capsnap, | |
46 | * eventually reaching the "live" or "head" context (i.e., pages that | |
47 | * are not yet snapped) and are writing the most recently dirtied | |
48 | * pages. | |
49 | * | |
50 | * Invalidate and so forth must take care to ensure the dirty page | |
51 | * accounting is preserved. | |
52 | */ | |
53 | ||
2baba250 YS |
54 | #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10)) |
55 | #define CONGESTION_OFF_THRESH(congestion_kb) \ | |
56 | (CONGESTION_ON_THRESH(congestion_kb) - \ | |
57 | (CONGESTION_ON_THRESH(congestion_kb) >> 2)) | |
58 | ||
61600ef8 YZ |
59 | static inline struct ceph_snap_context *page_snap_context(struct page *page) |
60 | { | |
61 | if (PagePrivate(page)) | |
62 | return (void *)page->private; | |
63 | return NULL; | |
64 | } | |
1d3576fd SW |
65 | |
66 | /* | |
67 | * Dirty a page. Optimistically adjust accounting, on the assumption | |
68 | * that we won't race with invalidate. If we do, readjust. | |
69 | */ | |
70 | static int ceph_set_page_dirty(struct page *page) | |
71 | { | |
72 | struct address_space *mapping = page->mapping; | |
73 | struct inode *inode; | |
74 | struct ceph_inode_info *ci; | |
1d3576fd | 75 | struct ceph_snap_context *snapc; |
7d6e1f54 | 76 | int ret; |
1d3576fd SW |
77 | |
78 | if (unlikely(!mapping)) | |
79 | return !TestSetPageDirty(page); | |
80 | ||
7d6e1f54 | 81 | if (PageDirty(page)) { |
1d3576fd SW |
82 | dout("%p set_page_dirty %p idx %lu -- already dirty\n", |
83 | mapping->host, page, page->index); | |
7d6e1f54 | 84 | BUG_ON(!PagePrivate(page)); |
1d3576fd SW |
85 | return 0; |
86 | } | |
87 | ||
88 | inode = mapping->host; | |
89 | ci = ceph_inode(inode); | |
90 | ||
1d3576fd | 91 | /* dirty the head */ |
be655596 | 92 | spin_lock(&ci->i_ceph_lock); |
5dda377c YZ |
93 | BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference |
94 | if (__ceph_have_pending_cap_snap(ci)) { | |
95 | struct ceph_cap_snap *capsnap = | |
96 | list_last_entry(&ci->i_cap_snaps, | |
97 | struct ceph_cap_snap, | |
98 | ci_item); | |
99 | snapc = ceph_get_snap_context(capsnap->context); | |
100 | capsnap->dirty_pages++; | |
101 | } else { | |
102 | BUG_ON(!ci->i_head_snapc); | |
103 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
104 | ++ci->i_wrbuffer_ref_head; | |
105 | } | |
1d3576fd | 106 | if (ci->i_wrbuffer_ref == 0) |
0444d76a | 107 | ihold(inode); |
1d3576fd SW |
108 | ++ci->i_wrbuffer_ref; |
109 | dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d " | |
110 | "snapc %p seq %lld (%d snaps)\n", | |
111 | mapping->host, page, page->index, | |
112 | ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1, | |
113 | ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, | |
114 | snapc, snapc->seq, snapc->num_snaps); | |
be655596 | 115 | spin_unlock(&ci->i_ceph_lock); |
1d3576fd | 116 | |
7d6e1f54 SZ |
117 | /* |
118 | * Reference snap context in page->private. Also set | |
119 | * PagePrivate so that we get invalidatepage callback. | |
120 | */ | |
121 | BUG_ON(PagePrivate(page)); | |
122 | page->private = (unsigned long)snapc; | |
123 | SetPagePrivate(page); | |
1d3576fd | 124 | |
7d6e1f54 SZ |
125 | ret = __set_page_dirty_nobuffers(page); |
126 | WARN_ON(!PageLocked(page)); | |
127 | WARN_ON(!page->mapping); | |
1d3576fd | 128 | |
7d6e1f54 | 129 | return ret; |
1d3576fd SW |
130 | } |
131 | ||
132 | /* | |
133 | * If we are truncating the full page (i.e. offset == 0), adjust the | |
134 | * dirty page counters appropriately. Only called if there is private | |
135 | * data on the page. | |
136 | */ | |
d47992f8 LC |
137 | static void ceph_invalidatepage(struct page *page, unsigned int offset, |
138 | unsigned int length) | |
1d3576fd | 139 | { |
4ce1e9ad | 140 | struct inode *inode; |
1d3576fd | 141 | struct ceph_inode_info *ci; |
61600ef8 | 142 | struct ceph_snap_context *snapc = page_snap_context(page); |
1d3576fd | 143 | |
4ce1e9ad | 144 | inode = page->mapping->host; |
b150f5c1 MT |
145 | ci = ceph_inode(inode); |
146 | ||
09cbfeaf | 147 | if (offset != 0 || length != PAGE_SIZE) { |
b150f5c1 MT |
148 | dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n", |
149 | inode, page, page->index, offset, length); | |
150 | return; | |
151 | } | |
4ce1e9ad | 152 | |
99ccbd22 MT |
153 | ceph_invalidate_fscache_page(inode, page); |
154 | ||
155 | if (!PagePrivate(page)) | |
156 | return; | |
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 | ||
b150f5c1 | 166 | ClearPageChecked(page); |
1d3576fd | 167 | |
b150f5c1 MT |
168 | dout("%p invalidatepage %p idx %lu full dirty page\n", |
169 | inode, page, page->index); | |
170 | ||
171 | ceph_put_wrbuffer_cap_refs(ci, 1, snapc); | |
172 | ceph_put_snap_context(snapc); | |
173 | page->private = 0; | |
174 | ClearPagePrivate(page); | |
1d3576fd SW |
175 | } |
176 | ||
1d3576fd SW |
177 | static int ceph_releasepage(struct page *page, gfp_t g) |
178 | { | |
e55f1a18 N |
179 | dout("%p releasepage %p idx %lu (%sdirty)\n", page->mapping->host, |
180 | page, page->index, PageDirty(page) ? "" : "not "); | |
99ccbd22 MT |
181 | |
182 | /* Can we release the page from the cache? */ | |
183 | if (!ceph_release_fscache_page(page, g)) | |
184 | return 0; | |
185 | ||
186 | return !PagePrivate(page); | |
1d3576fd SW |
187 | } |
188 | ||
189 | /* | |
190 | * read a single page, without unlocking it. | |
191 | */ | |
dd2bc473 | 192 | static int ceph_do_readpage(struct file *filp, struct page *page) |
1d3576fd | 193 | { |
496ad9aa | 194 | struct inode *inode = file_inode(filp); |
1d3576fd | 195 | struct ceph_inode_info *ci = ceph_inode(inode); |
99ccbd22 | 196 | struct ceph_osd_client *osdc = |
3d14c5d2 | 197 | &ceph_inode_to_client(inode)->client->osdc; |
1d3576fd | 198 | int err = 0; |
83701246 | 199 | u64 off = page_offset(page); |
09cbfeaf | 200 | u64 len = PAGE_SIZE; |
1d3576fd | 201 | |
83701246 | 202 | if (off >= i_size_read(inode)) { |
09cbfeaf | 203 | zero_user_segment(page, 0, PAGE_SIZE); |
83701246 YZ |
204 | SetPageUptodate(page); |
205 | return 0; | |
206 | } | |
99ccbd22 | 207 | |
fcc02d2a YZ |
208 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
209 | /* | |
210 | * Uptodate inline data should have been added | |
211 | * into page cache while getting Fcr caps. | |
212 | */ | |
213 | if (off == 0) | |
214 | return -EINVAL; | |
09cbfeaf | 215 | zero_user_segment(page, 0, PAGE_SIZE); |
fcc02d2a YZ |
216 | SetPageUptodate(page); |
217 | return 0; | |
218 | } | |
83701246 YZ |
219 | |
220 | err = ceph_readpage_from_fscache(inode, page); | |
99ccbd22 | 221 | if (err == 0) |
dd2bc473 | 222 | return -EINPROGRESS; |
99ccbd22 | 223 | |
1d3576fd SW |
224 | dout("readpage inode %p file %p page %p index %lu\n", |
225 | inode, filp, page, page->index); | |
226 | err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout, | |
83701246 | 227 | off, &len, |
1d3576fd | 228 | ci->i_truncate_seq, ci->i_truncate_size, |
b7495fc2 | 229 | &page, 1, 0); |
1d3576fd SW |
230 | if (err == -ENOENT) |
231 | err = 0; | |
232 | if (err < 0) { | |
233 | SetPageError(page); | |
18302805 | 234 | ceph_fscache_readpage_cancel(inode, page); |
1d3576fd | 235 | goto out; |
1d3576fd | 236 | } |
09cbfeaf | 237 | if (err < PAGE_SIZE) |
23cd573b | 238 | /* zero fill remainder of page */ |
09cbfeaf | 239 | zero_user_segment(page, err, PAGE_SIZE); |
23cd573b ZZ |
240 | else |
241 | flush_dcache_page(page); | |
1d3576fd | 242 | |
23cd573b ZZ |
243 | SetPageUptodate(page); |
244 | ceph_readpage_to_fscache(inode, page); | |
99ccbd22 | 245 | |
1d3576fd SW |
246 | out: |
247 | return err < 0 ? err : 0; | |
248 | } | |
249 | ||
250 | static int ceph_readpage(struct file *filp, struct page *page) | |
251 | { | |
dd2bc473 YZ |
252 | int r = ceph_do_readpage(filp, page); |
253 | if (r != -EINPROGRESS) | |
254 | unlock_page(page); | |
255 | else | |
256 | r = 0; | |
1d3576fd SW |
257 | return r; |
258 | } | |
259 | ||
260 | /* | |
7c272194 | 261 | * Finish an async read(ahead) op. |
1d3576fd | 262 | */ |
85e084fe | 263 | static void finish_read(struct ceph_osd_request *req) |
1d3576fd | 264 | { |
7c272194 | 265 | struct inode *inode = req->r_inode; |
87060c10 | 266 | struct ceph_osd_data *osd_data; |
85e084fe ID |
267 | int rc = req->r_result <= 0 ? req->r_result : 0; |
268 | int bytes = req->r_result >= 0 ? req->r_result : 0; | |
e0c59487 | 269 | int num_pages; |
7c272194 | 270 | int i; |
1d3576fd | 271 | |
7c272194 SW |
272 | dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes); |
273 | ||
274 | /* unlock all pages, zeroing any data we didn't read */ | |
406e2c9f | 275 | osd_data = osd_req_op_extent_osd_data(req, 0); |
87060c10 AE |
276 | BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); |
277 | num_pages = calc_pages_for((u64)osd_data->alignment, | |
278 | (u64)osd_data->length); | |
e0c59487 | 279 | for (i = 0; i < num_pages; i++) { |
87060c10 | 280 | struct page *page = osd_data->pages[i]; |
7c272194 | 281 | |
368e3585 YZ |
282 | if (rc < 0 && rc != -ENOENT) { |
283 | ceph_fscache_readpage_cancel(inode, page); | |
f36132a7 | 284 | goto unlock; |
368e3585 | 285 | } |
09cbfeaf | 286 | if (bytes < (int)PAGE_SIZE) { |
7c272194 SW |
287 | /* zero (remainder of) page */ |
288 | int s = bytes < 0 ? 0 : bytes; | |
09cbfeaf | 289 | zero_user_segment(page, s, PAGE_SIZE); |
1d3576fd | 290 | } |
7c272194 SW |
291 | dout("finish_read %p uptodate %p idx %lu\n", inode, page, |
292 | page->index); | |
293 | flush_dcache_page(page); | |
294 | SetPageUptodate(page); | |
99ccbd22 | 295 | ceph_readpage_to_fscache(inode, page); |
f36132a7 | 296 | unlock: |
7c272194 | 297 | unlock_page(page); |
09cbfeaf KS |
298 | put_page(page); |
299 | bytes -= PAGE_SIZE; | |
1d3576fd | 300 | } |
87060c10 | 301 | kfree(osd_data->pages); |
1d3576fd SW |
302 | } |
303 | ||
304 | /* | |
7c272194 SW |
305 | * start an async read(ahead) operation. return nr_pages we submitted |
306 | * a read for on success, or negative error code. | |
1d3576fd | 307 | */ |
0d66a487 | 308 | static int start_read(struct inode *inode, struct list_head *page_list, int max) |
1d3576fd | 309 | { |
3d14c5d2 YS |
310 | struct ceph_osd_client *osdc = |
311 | &ceph_inode_to_client(inode)->client->osdc; | |
7c272194 SW |
312 | struct ceph_inode_info *ci = ceph_inode(inode); |
313 | struct page *page = list_entry(page_list->prev, struct page, lru); | |
acead002 | 314 | struct ceph_vino vino; |
7c272194 SW |
315 | struct ceph_osd_request *req; |
316 | u64 off; | |
1d3576fd | 317 | u64 len; |
7c272194 SW |
318 | int i; |
319 | struct page **pages; | |
320 | pgoff_t next_index; | |
321 | int nr_pages = 0; | |
2b1ac852 YZ |
322 | int got = 0; |
323 | int ret = 0; | |
324 | ||
325 | if (!current->journal_info) { | |
326 | /* caller of readpages does not hold buffer and read caps | |
327 | * (fadvise, madvise and readahead cases) */ | |
328 | int want = CEPH_CAP_FILE_CACHE; | |
329 | ret = ceph_try_get_caps(ci, CEPH_CAP_FILE_RD, want, &got); | |
330 | if (ret < 0) { | |
331 | dout("start_read %p, error getting cap\n", inode); | |
332 | } else if (!(got & want)) { | |
333 | dout("start_read %p, no cache cap\n", inode); | |
334 | ret = 0; | |
335 | } | |
336 | if (ret <= 0) { | |
337 | if (got) | |
338 | ceph_put_cap_refs(ci, got); | |
339 | while (!list_empty(page_list)) { | |
340 | page = list_entry(page_list->prev, | |
341 | struct page, lru); | |
342 | list_del(&page->lru); | |
343 | put_page(page); | |
344 | } | |
345 | return ret; | |
346 | } | |
347 | } | |
1d3576fd | 348 | |
6285bc23 | 349 | off = (u64) page_offset(page); |
1d3576fd | 350 | |
7c272194 SW |
351 | /* count pages */ |
352 | next_index = page->index; | |
353 | list_for_each_entry_reverse(page, page_list, lru) { | |
354 | if (page->index != next_index) | |
355 | break; | |
356 | nr_pages++; | |
357 | next_index++; | |
0d66a487 SW |
358 | if (max && nr_pages == max) |
359 | break; | |
7c272194 | 360 | } |
09cbfeaf | 361 | len = nr_pages << PAGE_SHIFT; |
7c272194 SW |
362 | dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages, |
363 | off, len); | |
acead002 AE |
364 | vino = ceph_vino(inode); |
365 | req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len, | |
715e4cd4 | 366 | 0, 1, CEPH_OSD_OP_READ, |
acead002 | 367 | CEPH_OSD_FLAG_READ, NULL, |
7c272194 | 368 | ci->i_truncate_seq, ci->i_truncate_size, |
acead002 | 369 | false); |
2b1ac852 YZ |
370 | if (IS_ERR(req)) { |
371 | ret = PTR_ERR(req); | |
372 | goto out; | |
373 | } | |
1d3576fd | 374 | |
7c272194 | 375 | /* build page vector */ |
cf7b7e14 | 376 | nr_pages = calc_pages_for(0, len); |
687265e5 | 377 | pages = kmalloc(sizeof(*pages) * nr_pages, GFP_KERNEL); |
2b1ac852 YZ |
378 | if (!pages) { |
379 | ret = -ENOMEM; | |
380 | goto out_put; | |
381 | } | |
7c272194 SW |
382 | for (i = 0; i < nr_pages; ++i) { |
383 | page = list_entry(page_list->prev, struct page, lru); | |
384 | BUG_ON(PageLocked(page)); | |
1d3576fd | 385 | list_del(&page->lru); |
99ccbd22 | 386 | |
7c272194 SW |
387 | dout("start_read %p adding %p idx %lu\n", inode, page, |
388 | page->index); | |
389 | if (add_to_page_cache_lru(page, &inode->i_data, page->index, | |
687265e5 | 390 | GFP_KERNEL)) { |
d4d3aa38 | 391 | ceph_fscache_uncache_page(inode, page); |
09cbfeaf | 392 | put_page(page); |
7c272194 | 393 | dout("start_read %p add_to_page_cache failed %p\n", |
1d3576fd | 394 | inode, page); |
7c272194 | 395 | nr_pages = i; |
1afe4785 YZ |
396 | if (nr_pages > 0) { |
397 | len = nr_pages << PAGE_SHIFT; | |
d641df81 | 398 | osd_req_op_extent_update(req, 0, len); |
1afe4785 YZ |
399 | break; |
400 | } | |
7c272194 | 401 | goto out_pages; |
1d3576fd | 402 | } |
7c272194 | 403 | pages[i] = page; |
1d3576fd | 404 | } |
406e2c9f | 405 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false); |
7c272194 SW |
406 | req->r_callback = finish_read; |
407 | req->r_inode = inode; | |
408 | ||
409 | dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len); | |
410 | ret = ceph_osdc_start_request(osdc, req, false); | |
411 | if (ret < 0) | |
412 | goto out_pages; | |
413 | ceph_osdc_put_request(req); | |
2b1ac852 YZ |
414 | |
415 | /* After adding locked pages to page cache, the inode holds cache cap. | |
416 | * So we can drop our cap refs. */ | |
417 | if (got) | |
418 | ceph_put_cap_refs(ci, got); | |
419 | ||
7c272194 SW |
420 | return nr_pages; |
421 | ||
422 | out_pages: | |
1afe4785 YZ |
423 | for (i = 0; i < nr_pages; ++i) { |
424 | ceph_fscache_readpage_cancel(inode, pages[i]); | |
425 | unlock_page(pages[i]); | |
426 | } | |
427 | ceph_put_page_vector(pages, nr_pages, false); | |
2b1ac852 | 428 | out_put: |
7c272194 | 429 | ceph_osdc_put_request(req); |
2b1ac852 YZ |
430 | out: |
431 | if (got) | |
432 | ceph_put_cap_refs(ci, got); | |
7c272194 SW |
433 | return ret; |
434 | } | |
1d3576fd | 435 | |
7c272194 SW |
436 | |
437 | /* | |
438 | * Read multiple pages. Leave pages we don't read + unlock in page_list; | |
439 | * the caller (VM) cleans them up. | |
440 | */ | |
441 | static int ceph_readpages(struct file *file, struct address_space *mapping, | |
442 | struct list_head *page_list, unsigned nr_pages) | |
443 | { | |
496ad9aa | 444 | struct inode *inode = file_inode(file); |
0d66a487 | 445 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
7c272194 | 446 | int rc = 0; |
0d66a487 SW |
447 | int max = 0; |
448 | ||
83701246 YZ |
449 | if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE) |
450 | return -EINVAL; | |
451 | ||
99ccbd22 MT |
452 | rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list, |
453 | &nr_pages); | |
454 | ||
455 | if (rc == 0) | |
456 | goto out; | |
457 | ||
aa187926 YZ |
458 | max = fsc->mount_options->rsize >> PAGE_SHIFT; |
459 | dout("readpages %p file %p nr_pages %d max %d\n", | |
460 | inode, file, nr_pages, max); | |
7c272194 | 461 | while (!list_empty(page_list)) { |
0d66a487 | 462 | rc = start_read(inode, page_list, max); |
7c272194 SW |
463 | if (rc < 0) |
464 | goto out; | |
7c272194 | 465 | } |
1d3576fd | 466 | out: |
76be778b MT |
467 | ceph_fscache_readpages_cancel(inode, page_list); |
468 | ||
7c272194 | 469 | dout("readpages %p file %p ret %d\n", inode, file, rc); |
1d3576fd SW |
470 | return rc; |
471 | } | |
472 | ||
473 | /* | |
474 | * Get ref for the oldest snapc for an inode with dirty data... that is, the | |
475 | * only snap context we are allowed to write back. | |
1d3576fd | 476 | */ |
6298a337 | 477 | static struct ceph_snap_context *get_oldest_context(struct inode *inode, |
5f743e45 YZ |
478 | loff_t *snap_size, |
479 | u64 *truncate_size, | |
480 | u32 *truncate_seq) | |
1d3576fd SW |
481 | { |
482 | struct ceph_inode_info *ci = ceph_inode(inode); | |
483 | struct ceph_snap_context *snapc = NULL; | |
484 | struct ceph_cap_snap *capsnap = NULL; | |
485 | ||
be655596 | 486 | spin_lock(&ci->i_ceph_lock); |
1d3576fd SW |
487 | list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { |
488 | dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap, | |
489 | capsnap->context, capsnap->dirty_pages); | |
490 | if (capsnap->dirty_pages) { | |
491 | snapc = ceph_get_snap_context(capsnap->context); | |
492 | if (snap_size) | |
493 | *snap_size = capsnap->size; | |
5f743e45 YZ |
494 | if (truncate_size) |
495 | *truncate_size = capsnap->truncate_size; | |
496 | if (truncate_seq) | |
497 | *truncate_seq = capsnap->truncate_seq; | |
1d3576fd SW |
498 | break; |
499 | } | |
500 | } | |
7d8cb26d | 501 | if (!snapc && ci->i_wrbuffer_ref_head) { |
80e755fe | 502 | snapc = ceph_get_snap_context(ci->i_head_snapc); |
1d3576fd SW |
503 | dout(" head snapc %p has %d dirty pages\n", |
504 | snapc, ci->i_wrbuffer_ref_head); | |
5f743e45 | 505 | if (truncate_size) |
84fcc2d2 | 506 | *truncate_size = ci->i_truncate_size; |
5f743e45 | 507 | if (truncate_seq) |
84fcc2d2 | 508 | *truncate_seq = ci->i_truncate_seq; |
1d3576fd | 509 | } |
be655596 | 510 | spin_unlock(&ci->i_ceph_lock); |
1d3576fd SW |
511 | return snapc; |
512 | } | |
513 | ||
514 | /* | |
515 | * Write a single page, but leave the page locked. | |
516 | * | |
517 | * If we get a write error, set the page error bit, but still adjust the | |
518 | * dirty page accounting (i.e., page is no longer dirty). | |
519 | */ | |
520 | static int writepage_nounlock(struct page *page, struct writeback_control *wbc) | |
521 | { | |
522 | struct inode *inode; | |
523 | struct ceph_inode_info *ci; | |
3d14c5d2 | 524 | struct ceph_fs_client *fsc; |
1d3576fd | 525 | struct ceph_osd_client *osdc; |
6298a337 | 526 | struct ceph_snap_context *snapc, *oldest; |
fc2744aa | 527 | loff_t page_off = page_offset(page); |
e1966b49 | 528 | loff_t snap_size = -1; |
2baba250 | 529 | long writeback_stat; |
e1966b49 | 530 | u64 truncate_size; |
fc2744aa | 531 | u32 truncate_seq; |
43986881 | 532 | int err, len = PAGE_SIZE; |
1d3576fd SW |
533 | |
534 | dout("writepage %p idx %lu\n", page, page->index); | |
535 | ||
1d3576fd SW |
536 | inode = page->mapping->host; |
537 | ci = ceph_inode(inode); | |
3d14c5d2 YS |
538 | fsc = ceph_inode_to_client(inode); |
539 | osdc = &fsc->client->osdc; | |
1d3576fd SW |
540 | |
541 | /* verify this is a writeable snap context */ | |
61600ef8 | 542 | snapc = page_snap_context(page); |
d37b1d99 | 543 | if (!snapc) { |
1d3576fd | 544 | dout("writepage %p page %p not dirty?\n", inode, page); |
43986881 | 545 | return 0; |
1d3576fd | 546 | } |
5f743e45 YZ |
547 | oldest = get_oldest_context(inode, &snap_size, |
548 | &truncate_size, &truncate_seq); | |
6298a337 | 549 | if (snapc->seq > oldest->seq) { |
1d3576fd | 550 | dout("writepage %p page %p snapc %p not writeable - noop\n", |
61600ef8 | 551 | inode, page, snapc); |
1d3576fd | 552 | /* we should only noop if called by kswapd */ |
fa71fefb | 553 | WARN_ON(!(current->flags & PF_MEMALLOC)); |
6298a337 | 554 | ceph_put_snap_context(oldest); |
fa71fefb | 555 | redirty_page_for_writepage(wbc, page); |
43986881 | 556 | return 0; |
1d3576fd | 557 | } |
6298a337 | 558 | ceph_put_snap_context(oldest); |
1d3576fd | 559 | |
e1966b49 | 560 | if (snap_size == -1) |
fc2744aa | 561 | snap_size = i_size_read(inode); |
fc2744aa | 562 | |
1d3576fd | 563 | /* is this a partial page at end of file? */ |
fc2744aa YZ |
564 | if (page_off >= snap_size) { |
565 | dout("%p page eof %llu\n", page, snap_size); | |
43986881 | 566 | return 0; |
fc2744aa | 567 | } |
43986881 | 568 | |
fc2744aa YZ |
569 | if (snap_size < page_off + len) |
570 | len = snap_size - page_off; | |
1d3576fd | 571 | |
1c0a9c2d YZ |
572 | dout("writepage %p page %p index %lu on %llu~%u snapc %p seq %lld\n", |
573 | inode, page, page->index, page_off, len, snapc, snapc->seq); | |
1d3576fd | 574 | |
3d14c5d2 | 575 | writeback_stat = atomic_long_inc_return(&fsc->writeback_count); |
2baba250 | 576 | if (writeback_stat > |
3d14c5d2 | 577 | CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb)) |
09dc9fc2 | 578 | set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); |
2baba250 | 579 | |
1d3576fd SW |
580 | set_page_writeback(page); |
581 | err = ceph_osdc_writepages(osdc, ceph_vino(inode), | |
582 | &ci->i_layout, snapc, | |
583 | page_off, len, | |
fc2744aa | 584 | truncate_seq, truncate_size, |
24808826 | 585 | &inode->i_mtime, &page, 1); |
1d3576fd | 586 | if (err < 0) { |
ad15ec06 YZ |
587 | struct writeback_control tmp_wbc; |
588 | if (!wbc) | |
589 | wbc = &tmp_wbc; | |
590 | if (err == -ERESTARTSYS) { | |
591 | /* killed by SIGKILL */ | |
592 | dout("writepage interrupted page %p\n", page); | |
593 | redirty_page_for_writepage(wbc, page); | |
594 | end_page_writeback(page); | |
43986881 | 595 | return err; |
ad15ec06 YZ |
596 | } |
597 | dout("writepage setting page/mapping error %d %p\n", | |
598 | err, page); | |
1d3576fd SW |
599 | SetPageError(page); |
600 | mapping_set_error(&inode->i_data, err); | |
ad15ec06 | 601 | wbc->pages_skipped++; |
1d3576fd SW |
602 | } else { |
603 | dout("writepage cleaned page %p\n", page); | |
604 | err = 0; /* vfs expects us to return 0 */ | |
605 | } | |
606 | page->private = 0; | |
607 | ClearPagePrivate(page); | |
608 | end_page_writeback(page); | |
609 | ceph_put_wrbuffer_cap_refs(ci, 1, snapc); | |
6298a337 | 610 | ceph_put_snap_context(snapc); /* page's reference */ |
1d3576fd SW |
611 | return err; |
612 | } | |
613 | ||
614 | static int ceph_writepage(struct page *page, struct writeback_control *wbc) | |
615 | { | |
dbd646a8 YS |
616 | int err; |
617 | struct inode *inode = page->mapping->host; | |
618 | BUG_ON(!inode); | |
70b666c3 | 619 | ihold(inode); |
dbd646a8 | 620 | err = writepage_nounlock(page, wbc); |
ad15ec06 YZ |
621 | if (err == -ERESTARTSYS) { |
622 | /* direct memory reclaimer was killed by SIGKILL. return 0 | |
623 | * to prevent caller from setting mapping/page error */ | |
624 | err = 0; | |
625 | } | |
1d3576fd | 626 | unlock_page(page); |
dbd646a8 | 627 | iput(inode); |
1d3576fd SW |
628 | return err; |
629 | } | |
630 | ||
1d3576fd SW |
631 | /* |
632 | * lame release_pages helper. release_pages() isn't exported to | |
633 | * modules. | |
634 | */ | |
635 | static void ceph_release_pages(struct page **pages, int num) | |
636 | { | |
637 | struct pagevec pvec; | |
638 | int i; | |
639 | ||
640 | pagevec_init(&pvec, 0); | |
641 | for (i = 0; i < num; i++) { | |
642 | if (pagevec_add(&pvec, pages[i]) == 0) | |
643 | pagevec_release(&pvec); | |
644 | } | |
645 | pagevec_release(&pvec); | |
646 | } | |
647 | ||
1d3576fd SW |
648 | /* |
649 | * async writeback completion handler. | |
650 | * | |
651 | * If we get an error, set the mapping error bit, but not the individual | |
652 | * page error bits. | |
653 | */ | |
85e084fe | 654 | static void writepages_finish(struct ceph_osd_request *req) |
1d3576fd SW |
655 | { |
656 | struct inode *inode = req->r_inode; | |
1d3576fd | 657 | struct ceph_inode_info *ci = ceph_inode(inode); |
87060c10 | 658 | struct ceph_osd_data *osd_data; |
1d3576fd | 659 | struct page *page; |
5b64640c YZ |
660 | int num_pages, total_pages = 0; |
661 | int i, j; | |
662 | int rc = req->r_result; | |
1d3576fd SW |
663 | struct ceph_snap_context *snapc = req->r_snapc; |
664 | struct address_space *mapping = inode->i_mapping; | |
3d14c5d2 | 665 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
5b64640c | 666 | bool remove_page; |
1d3576fd | 667 | |
5b64640c | 668 | dout("writepages_finish %p rc %d\n", inode, rc); |
26544c62 | 669 | if (rc < 0) { |
1d3576fd | 670 | mapping_set_error(mapping, rc); |
26544c62 JL |
671 | ceph_set_error_write(ci); |
672 | } else { | |
673 | ceph_clear_error_write(ci); | |
674 | } | |
5b64640c YZ |
675 | |
676 | /* | |
677 | * We lost the cache cap, need to truncate the page before | |
678 | * it is unlocked, otherwise we'd truncate it later in the | |
679 | * page truncation thread, possibly losing some data that | |
680 | * raced its way in | |
681 | */ | |
682 | remove_page = !(ceph_caps_issued(ci) & | |
683 | (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)); | |
1d3576fd SW |
684 | |
685 | /* clean all pages */ | |
5b64640c YZ |
686 | for (i = 0; i < req->r_num_ops; i++) { |
687 | if (req->r_ops[i].op != CEPH_OSD_OP_WRITE) | |
688 | break; | |
e63dc5c7 | 689 | |
5b64640c YZ |
690 | osd_data = osd_req_op_extent_osd_data(req, i); |
691 | BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); | |
692 | num_pages = calc_pages_for((u64)osd_data->alignment, | |
693 | (u64)osd_data->length); | |
694 | total_pages += num_pages; | |
695 | for (j = 0; j < num_pages; j++) { | |
696 | page = osd_data->pages[j]; | |
697 | BUG_ON(!page); | |
698 | WARN_ON(!PageUptodate(page)); | |
699 | ||
700 | if (atomic_long_dec_return(&fsc->writeback_count) < | |
701 | CONGESTION_OFF_THRESH( | |
702 | fsc->mount_options->congestion_kb)) | |
09dc9fc2 | 703 | clear_bdi_congested(inode_to_bdi(inode), |
5b64640c YZ |
704 | BLK_RW_ASYNC); |
705 | ||
706 | ceph_put_snap_context(page_snap_context(page)); | |
707 | page->private = 0; | |
708 | ClearPagePrivate(page); | |
709 | dout("unlocking %p\n", page); | |
710 | end_page_writeback(page); | |
711 | ||
712 | if (remove_page) | |
713 | generic_error_remove_page(inode->i_mapping, | |
714 | page); | |
715 | ||
716 | unlock_page(page); | |
717 | } | |
718 | dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n", | |
719 | inode, osd_data->length, rc >= 0 ? num_pages : 0); | |
e63dc5c7 | 720 | |
5b64640c | 721 | ceph_release_pages(osd_data->pages, num_pages); |
1d3576fd | 722 | } |
1d3576fd | 723 | |
5b64640c YZ |
724 | ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc); |
725 | ||
726 | osd_data = osd_req_op_extent_osd_data(req, 0); | |
87060c10 AE |
727 | if (osd_data->pages_from_pool) |
728 | mempool_free(osd_data->pages, | |
640ef79d | 729 | ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool); |
1d3576fd | 730 | else |
87060c10 | 731 | kfree(osd_data->pages); |
1d3576fd SW |
732 | ceph_osdc_put_request(req); |
733 | } | |
734 | ||
1d3576fd SW |
735 | /* |
736 | * initiate async writeback | |
737 | */ | |
738 | static int ceph_writepages_start(struct address_space *mapping, | |
739 | struct writeback_control *wbc) | |
740 | { | |
741 | struct inode *inode = mapping->host; | |
1d3576fd | 742 | struct ceph_inode_info *ci = ceph_inode(inode); |
fc2744aa YZ |
743 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
744 | struct ceph_vino vino = ceph_vino(inode); | |
1d3576fd SW |
745 | pgoff_t index, start, end; |
746 | int range_whole = 0; | |
747 | int should_loop = 1; | |
748 | pgoff_t max_pages = 0, max_pages_ever = 0; | |
80e755fe | 749 | struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc; |
1d3576fd SW |
750 | struct pagevec pvec; |
751 | int done = 0; | |
752 | int rc = 0; | |
93407472 | 753 | unsigned int wsize = i_blocksize(inode); |
1d3576fd | 754 | struct ceph_osd_request *req = NULL; |
e1966b49 YZ |
755 | loff_t snap_size, i_size; |
756 | u64 truncate_size; | |
fc2744aa | 757 | u32 truncate_seq; |
1d3576fd | 758 | |
3fb99d48 | 759 | dout("writepages_start %p (mode=%s)\n", inode, |
1d3576fd SW |
760 | wbc->sync_mode == WB_SYNC_NONE ? "NONE" : |
761 | (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD")); | |
762 | ||
52953d55 | 763 | if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { |
6c93df5d YZ |
764 | if (ci->i_wrbuffer_ref > 0) { |
765 | pr_warn_ratelimited( | |
766 | "writepage_start %p %lld forced umount\n", | |
767 | inode, ceph_ino(inode)); | |
768 | } | |
a341d4df | 769 | mapping_set_error(mapping, -EIO); |
1d3576fd SW |
770 | return -EIO; /* we're in a forced umount, don't write! */ |
771 | } | |
95cca2b4 | 772 | if (fsc->mount_options->wsize < wsize) |
3d14c5d2 | 773 | wsize = fsc->mount_options->wsize; |
09cbfeaf | 774 | max_pages_ever = wsize >> PAGE_SHIFT; |
1d3576fd SW |
775 | |
776 | pagevec_init(&pvec, 0); | |
777 | ||
1d3576fd SW |
778 | /* where to start/end? */ |
779 | if (wbc->range_cyclic) { | |
780 | start = mapping->writeback_index; /* Start from prev offset */ | |
781 | end = -1; | |
782 | dout(" cyclic, start at %lu\n", start); | |
783 | } else { | |
09cbfeaf KS |
784 | start = wbc->range_start >> PAGE_SHIFT; |
785 | end = wbc->range_end >> PAGE_SHIFT; | |
1d3576fd SW |
786 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
787 | range_whole = 1; | |
788 | should_loop = 0; | |
789 | dout(" not cyclic, %lu to %lu\n", start, end); | |
790 | } | |
791 | index = start; | |
792 | ||
793 | retry: | |
794 | /* find oldest snap context with dirty data */ | |
795 | ceph_put_snap_context(snapc); | |
e1966b49 | 796 | snap_size = -1; |
5f743e45 YZ |
797 | snapc = get_oldest_context(inode, &snap_size, |
798 | &truncate_size, &truncate_seq); | |
1d3576fd SW |
799 | if (!snapc) { |
800 | /* hmm, why does writepages get called when there | |
801 | is no dirty data? */ | |
802 | dout(" no snap context with dirty data?\n"); | |
803 | goto out; | |
804 | } | |
805 | dout(" oldest snapc is %p seq %lld (%d snaps)\n", | |
806 | snapc, snapc->seq, snapc->num_snaps); | |
fc2744aa | 807 | |
e1966b49 | 808 | i_size = i_size_read(inode); |
fc2744aa | 809 | |
1d3576fd SW |
810 | if (last_snapc && snapc != last_snapc) { |
811 | /* if we switched to a newer snapc, restart our scan at the | |
812 | * start of the original file range. */ | |
813 | dout(" snapc differs from last pass, restarting at %lu\n", | |
814 | index); | |
815 | index = start; | |
816 | } | |
817 | last_snapc = snapc; | |
818 | ||
819 | while (!done && index <= end) { | |
820 | unsigned i; | |
821 | int first; | |
5b64640c YZ |
822 | pgoff_t strip_unit_end = 0; |
823 | int num_ops = 0, op_idx; | |
824 | int pvec_pages, locked_pages = 0; | |
825 | struct page **pages = NULL, **data_pages; | |
e5975c7c | 826 | mempool_t *pool = NULL; /* Becomes non-null if mempool used */ |
1d3576fd SW |
827 | struct page *page; |
828 | int want; | |
5b64640c | 829 | u64 offset = 0, len = 0; |
1d3576fd | 830 | |
1d3576fd SW |
831 | max_pages = max_pages_ever; |
832 | ||
833 | get_more_pages: | |
834 | first = -1; | |
835 | want = min(end - index, | |
836 | min((pgoff_t)PAGEVEC_SIZE, | |
837 | max_pages - (pgoff_t)locked_pages) - 1) | |
838 | + 1; | |
839 | pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
840 | PAGECACHE_TAG_DIRTY, | |
841 | want); | |
842 | dout("pagevec_lookup_tag got %d\n", pvec_pages); | |
843 | if (!pvec_pages && !locked_pages) | |
844 | break; | |
845 | for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) { | |
846 | page = pvec.pages[i]; | |
847 | dout("? %p idx %lu\n", page, page->index); | |
848 | if (locked_pages == 0) | |
849 | lock_page(page); /* first page */ | |
850 | else if (!trylock_page(page)) | |
851 | break; | |
852 | ||
853 | /* only dirty pages, or our accounting breaks */ | |
854 | if (unlikely(!PageDirty(page)) || | |
855 | unlikely(page->mapping != mapping)) { | |
856 | dout("!dirty or !mapping %p\n", page); | |
857 | unlock_page(page); | |
858 | break; | |
859 | } | |
860 | if (!wbc->range_cyclic && page->index > end) { | |
861 | dout("end of range %p\n", page); | |
862 | done = 1; | |
863 | unlock_page(page); | |
864 | break; | |
865 | } | |
5b64640c YZ |
866 | if (strip_unit_end && (page->index > strip_unit_end)) { |
867 | dout("end of strip unit %p\n", page); | |
1d3576fd SW |
868 | unlock_page(page); |
869 | break; | |
870 | } | |
871 | if (wbc->sync_mode != WB_SYNC_NONE) { | |
872 | dout("waiting on writeback %p\n", page); | |
873 | wait_on_page_writeback(page); | |
874 | } | |
e1966b49 YZ |
875 | if (page_offset(page) >= |
876 | (snap_size == -1 ? i_size : snap_size)) { | |
877 | dout("%p page eof %llu\n", page, | |
878 | (snap_size == -1 ? i_size : snap_size)); | |
1d3576fd SW |
879 | done = 1; |
880 | unlock_page(page); | |
881 | break; | |
882 | } | |
883 | if (PageWriteback(page)) { | |
884 | dout("%p under writeback\n", page); | |
885 | unlock_page(page); | |
886 | break; | |
887 | } | |
888 | ||
889 | /* only if matching snap context */ | |
61600ef8 | 890 | pgsnapc = page_snap_context(page); |
80e755fe SW |
891 | if (pgsnapc->seq > snapc->seq) { |
892 | dout("page snapc %p %lld > oldest %p %lld\n", | |
893 | pgsnapc, pgsnapc->seq, snapc, snapc->seq); | |
1d3576fd SW |
894 | unlock_page(page); |
895 | if (!locked_pages) | |
896 | continue; /* keep looking for snap */ | |
897 | break; | |
898 | } | |
899 | ||
900 | if (!clear_page_dirty_for_io(page)) { | |
901 | dout("%p !clear_page_dirty_for_io\n", page); | |
902 | unlock_page(page); | |
903 | break; | |
904 | } | |
905 | ||
e5975c7c AE |
906 | /* |
907 | * We have something to write. If this is | |
908 | * the first locked page this time through, | |
5b64640c YZ |
909 | * calculate max possinle write size and |
910 | * allocate a page array | |
e5975c7c | 911 | */ |
1d3576fd | 912 | if (locked_pages == 0) { |
5b64640c YZ |
913 | u64 objnum; |
914 | u64 objoff; | |
915 | ||
1d3576fd | 916 | /* prepare async write request */ |
e5975c7c | 917 | offset = (u64)page_offset(page); |
1d3576fd | 918 | len = wsize; |
5b64640c YZ |
919 | |
920 | rc = ceph_calc_file_object_mapping(&ci->i_layout, | |
921 | offset, len, | |
922 | &objnum, &objoff, | |
923 | &len); | |
924 | if (rc < 0) { | |
8c71897b HC |
925 | unlock_page(page); |
926 | break; | |
927 | } | |
928 | ||
3fb99d48 | 929 | num_ops = 1; |
5b64640c | 930 | strip_unit_end = page->index + |
09cbfeaf | 931 | ((len - 1) >> PAGE_SHIFT); |
88486957 | 932 | |
5b64640c | 933 | BUG_ON(pages); |
88486957 | 934 | max_pages = calc_pages_for(0, (u64)len); |
fc2744aa YZ |
935 | pages = kmalloc(max_pages * sizeof (*pages), |
936 | GFP_NOFS); | |
88486957 AE |
937 | if (!pages) { |
938 | pool = fsc->wb_pagevec_pool; | |
88486957 | 939 | pages = mempool_alloc(pool, GFP_NOFS); |
e5975c7c | 940 | BUG_ON(!pages); |
88486957 | 941 | } |
5b64640c YZ |
942 | |
943 | len = 0; | |
944 | } else if (page->index != | |
09cbfeaf | 945 | (offset + len) >> PAGE_SHIFT) { |
5b64640c YZ |
946 | if (num_ops >= (pool ? CEPH_OSD_SLAB_OPS : |
947 | CEPH_OSD_MAX_OPS)) { | |
948 | redirty_page_for_writepage(wbc, page); | |
949 | unlock_page(page); | |
950 | break; | |
951 | } | |
952 | ||
953 | num_ops++; | |
954 | offset = (u64)page_offset(page); | |
955 | len = 0; | |
1d3576fd SW |
956 | } |
957 | ||
958 | /* note position of first page in pvec */ | |
959 | if (first < 0) | |
960 | first = i; | |
961 | dout("%p will write page %p idx %lu\n", | |
962 | inode, page, page->index); | |
2baba250 | 963 | |
5b64640c YZ |
964 | if (atomic_long_inc_return(&fsc->writeback_count) > |
965 | CONGESTION_ON_THRESH( | |
3d14c5d2 | 966 | fsc->mount_options->congestion_kb)) { |
09dc9fc2 | 967 | set_bdi_congested(inode_to_bdi(inode), |
213c99ee | 968 | BLK_RW_ASYNC); |
2baba250 YS |
969 | } |
970 | ||
e5975c7c | 971 | pages[locked_pages] = page; |
1d3576fd | 972 | locked_pages++; |
09cbfeaf | 973 | len += PAGE_SIZE; |
1d3576fd SW |
974 | } |
975 | ||
976 | /* did we get anything? */ | |
977 | if (!locked_pages) | |
978 | goto release_pvec_pages; | |
979 | if (i) { | |
980 | int j; | |
981 | BUG_ON(!locked_pages || first < 0); | |
982 | ||
983 | if (pvec_pages && i == pvec_pages && | |
984 | locked_pages < max_pages) { | |
985 | dout("reached end pvec, trying for more\n"); | |
986 | pagevec_reinit(&pvec); | |
987 | goto get_more_pages; | |
988 | } | |
989 | ||
990 | /* shift unused pages over in the pvec... we | |
991 | * will need to release them below. */ | |
992 | for (j = i; j < pvec_pages; j++) { | |
5b64640c | 993 | dout(" pvec leftover page %p\n", pvec.pages[j]); |
1d3576fd SW |
994 | pvec.pages[j-i+first] = pvec.pages[j]; |
995 | } | |
996 | pvec.nr -= i-first; | |
997 | } | |
998 | ||
5b64640c | 999 | new_request: |
e5975c7c | 1000 | offset = page_offset(pages[0]); |
5b64640c YZ |
1001 | len = wsize; |
1002 | ||
1003 | req = ceph_osdc_new_request(&fsc->client->osdc, | |
1004 | &ci->i_layout, vino, | |
1005 | offset, &len, 0, num_ops, | |
1006 | CEPH_OSD_OP_WRITE, | |
54ea0046 | 1007 | CEPH_OSD_FLAG_WRITE, |
5b64640c YZ |
1008 | snapc, truncate_seq, |
1009 | truncate_size, false); | |
1010 | if (IS_ERR(req)) { | |
1011 | req = ceph_osdc_new_request(&fsc->client->osdc, | |
1012 | &ci->i_layout, vino, | |
1013 | offset, &len, 0, | |
1014 | min(num_ops, | |
1015 | CEPH_OSD_SLAB_OPS), | |
1016 | CEPH_OSD_OP_WRITE, | |
54ea0046 | 1017 | CEPH_OSD_FLAG_WRITE, |
5b64640c YZ |
1018 | snapc, truncate_seq, |
1019 | truncate_size, true); | |
1020 | BUG_ON(IS_ERR(req)); | |
e1966b49 | 1021 | } |
5b64640c | 1022 | BUG_ON(len < page_offset(pages[locked_pages - 1]) + |
09cbfeaf | 1023 | PAGE_SIZE - offset); |
5b64640c YZ |
1024 | |
1025 | req->r_callback = writepages_finish; | |
1026 | req->r_inode = inode; | |
1d3576fd | 1027 | |
5b64640c YZ |
1028 | /* Format the osd request message and submit the write */ |
1029 | len = 0; | |
1030 | data_pages = pages; | |
1031 | op_idx = 0; | |
1032 | for (i = 0; i < locked_pages; i++) { | |
1033 | u64 cur_offset = page_offset(pages[i]); | |
1034 | if (offset + len != cur_offset) { | |
3fb99d48 | 1035 | if (op_idx + 1 == req->r_num_ops) |
5b64640c YZ |
1036 | break; |
1037 | osd_req_op_extent_dup_last(req, op_idx, | |
1038 | cur_offset - offset); | |
1039 | dout("writepages got pages at %llu~%llu\n", | |
1040 | offset, len); | |
1041 | osd_req_op_extent_osd_data_pages(req, op_idx, | |
1042 | data_pages, len, 0, | |
a4ce40a9 | 1043 | !!pool, false); |
5b64640c | 1044 | osd_req_op_extent_update(req, op_idx, len); |
e5975c7c | 1045 | |
5b64640c YZ |
1046 | len = 0; |
1047 | offset = cur_offset; | |
1048 | data_pages = pages + i; | |
1049 | op_idx++; | |
1050 | } | |
1051 | ||
1052 | set_page_writeback(pages[i]); | |
09cbfeaf | 1053 | len += PAGE_SIZE; |
5b64640c YZ |
1054 | } |
1055 | ||
1056 | if (snap_size != -1) { | |
1057 | len = min(len, snap_size - offset); | |
1058 | } else if (i == locked_pages) { | |
1059 | /* writepages_finish() clears writeback pages | |
1060 | * according to the data length, so make sure | |
1061 | * data length covers all locked pages */ | |
09cbfeaf | 1062 | u64 min_len = len + 1 - PAGE_SIZE; |
5b64640c YZ |
1063 | len = min(len, (u64)i_size_read(inode) - offset); |
1064 | len = max(len, min_len); | |
1065 | } | |
1066 | dout("writepages got pages at %llu~%llu\n", offset, len); | |
e5975c7c | 1067 | |
5b64640c YZ |
1068 | osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len, |
1069 | 0, !!pool, false); | |
1070 | osd_req_op_extent_update(req, op_idx, len); | |
e5975c7c | 1071 | |
5b64640c YZ |
1072 | BUG_ON(op_idx + 1 != req->r_num_ops); |
1073 | ||
1074 | pool = NULL; | |
1075 | if (i < locked_pages) { | |
1076 | BUG_ON(num_ops <= req->r_num_ops); | |
1077 | num_ops -= req->r_num_ops; | |
5b64640c YZ |
1078 | locked_pages -= i; |
1079 | ||
1080 | /* allocate new pages array for next request */ | |
1081 | data_pages = pages; | |
1082 | pages = kmalloc(locked_pages * sizeof (*pages), | |
1083 | GFP_NOFS); | |
1084 | if (!pages) { | |
1085 | pool = fsc->wb_pagevec_pool; | |
1086 | pages = mempool_alloc(pool, GFP_NOFS); | |
1087 | BUG_ON(!pages); | |
1088 | } | |
1089 | memcpy(pages, data_pages + i, | |
1090 | locked_pages * sizeof(*pages)); | |
1091 | memset(data_pages + i, 0, | |
1092 | locked_pages * sizeof(*pages)); | |
1093 | } else { | |
1094 | BUG_ON(num_ops != req->r_num_ops); | |
1095 | index = pages[i - 1]->index + 1; | |
1096 | /* request message now owns the pages array */ | |
1097 | pages = NULL; | |
1098 | } | |
e5975c7c | 1099 | |
bb873b53 | 1100 | req->r_mtime = inode->i_mtime; |
9d6fcb08 SW |
1101 | rc = ceph_osdc_start_request(&fsc->client->osdc, req, true); |
1102 | BUG_ON(rc); | |
1d3576fd SW |
1103 | req = NULL; |
1104 | ||
5b64640c YZ |
1105 | wbc->nr_to_write -= i; |
1106 | if (pages) | |
1107 | goto new_request; | |
1108 | ||
1d3576fd SW |
1109 | if (wbc->nr_to_write <= 0) |
1110 | done = 1; | |
1111 | ||
1112 | release_pvec_pages: | |
1113 | dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr, | |
1114 | pvec.nr ? pvec.pages[0] : NULL); | |
1115 | pagevec_release(&pvec); | |
1116 | ||
1117 | if (locked_pages && !done) | |
1118 | goto retry; | |
1119 | } | |
1120 | ||
1121 | if (should_loop && !done) { | |
1122 | /* more to do; loop back to beginning of file */ | |
1123 | dout("writepages looping back to beginning of file\n"); | |
1124 | should_loop = 0; | |
1125 | index = 0; | |
1126 | goto retry; | |
1127 | } | |
1128 | ||
1129 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) | |
1130 | mapping->writeback_index = index; | |
1131 | ||
1132 | out: | |
3ed97d63 | 1133 | ceph_osdc_put_request(req); |
1d3576fd SW |
1134 | ceph_put_snap_context(snapc); |
1135 | dout("writepages done, rc = %d\n", rc); | |
1d3576fd SW |
1136 | return rc; |
1137 | } | |
1138 | ||
1139 | ||
1140 | ||
1141 | /* | |
1142 | * See if a given @snapc is either writeable, or already written. | |
1143 | */ | |
1144 | static int context_is_writeable_or_written(struct inode *inode, | |
1145 | struct ceph_snap_context *snapc) | |
1146 | { | |
5f743e45 YZ |
1147 | struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, |
1148 | NULL, NULL); | |
6298a337 SW |
1149 | int ret = !oldest || snapc->seq <= oldest->seq; |
1150 | ||
1151 | ceph_put_snap_context(oldest); | |
1152 | return ret; | |
1d3576fd SW |
1153 | } |
1154 | ||
1155 | /* | |
1156 | * We are only allowed to write into/dirty the page if the page is | |
1157 | * clean, or already dirty within the same snap context. | |
8f883c24 SW |
1158 | * |
1159 | * called with page locked. | |
1160 | * return success with page locked, | |
1161 | * or any failure (incl -EAGAIN) with page unlocked. | |
1d3576fd | 1162 | */ |
4af6b225 YS |
1163 | static int ceph_update_writeable_page(struct file *file, |
1164 | loff_t pos, unsigned len, | |
1165 | struct page *page) | |
1d3576fd | 1166 | { |
496ad9aa | 1167 | struct inode *inode = file_inode(file); |
6c93df5d | 1168 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
1d3576fd | 1169 | struct ceph_inode_info *ci = ceph_inode(inode); |
09cbfeaf KS |
1170 | loff_t page_off = pos & PAGE_MASK; |
1171 | int pos_in_page = pos & ~PAGE_MASK; | |
1d3576fd SW |
1172 | int end_in_page = pos_in_page + len; |
1173 | loff_t i_size; | |
1d3576fd | 1174 | int r; |
80e755fe | 1175 | struct ceph_snap_context *snapc, *oldest; |
1d3576fd | 1176 | |
52953d55 | 1177 | if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { |
6c93df5d YZ |
1178 | dout(" page %p forced umount\n", page); |
1179 | unlock_page(page); | |
1180 | return -EIO; | |
1181 | } | |
1182 | ||
1d3576fd SW |
1183 | retry_locked: |
1184 | /* writepages currently holds page lock, but if we change that later, */ | |
1185 | wait_on_page_writeback(page); | |
1186 | ||
61600ef8 | 1187 | snapc = page_snap_context(page); |
80e755fe | 1188 | if (snapc && snapc != ci->i_head_snapc) { |
1d3576fd SW |
1189 | /* |
1190 | * this page is already dirty in another (older) snap | |
1191 | * context! is it writeable now? | |
1192 | */ | |
5f743e45 | 1193 | oldest = get_oldest_context(inode, NULL, NULL, NULL); |
1d3576fd | 1194 | |
80e755fe | 1195 | if (snapc->seq > oldest->seq) { |
6298a337 | 1196 | ceph_put_snap_context(oldest); |
1d3576fd | 1197 | dout(" page %p snapc %p not current or oldest\n", |
6298a337 | 1198 | page, snapc); |
1d3576fd SW |
1199 | /* |
1200 | * queue for writeback, and wait for snapc to | |
1201 | * be writeable or written | |
1202 | */ | |
6298a337 | 1203 | snapc = ceph_get_snap_context(snapc); |
1d3576fd | 1204 | unlock_page(page); |
3c6f6b79 | 1205 | ceph_queue_writeback(inode); |
a78bbd4b | 1206 | r = wait_event_killable(ci->i_cap_wq, |
1d3576fd SW |
1207 | context_is_writeable_or_written(inode, snapc)); |
1208 | ceph_put_snap_context(snapc); | |
8f883c24 SW |
1209 | if (r == -ERESTARTSYS) |
1210 | return r; | |
4af6b225 | 1211 | return -EAGAIN; |
1d3576fd | 1212 | } |
6298a337 | 1213 | ceph_put_snap_context(oldest); |
1d3576fd SW |
1214 | |
1215 | /* yay, writeable, do it now (without dropping page lock) */ | |
1216 | dout(" page %p snapc %p not current, but oldest\n", | |
1217 | page, snapc); | |
1218 | if (!clear_page_dirty_for_io(page)) | |
1219 | goto retry_locked; | |
1220 | r = writepage_nounlock(page, NULL); | |
1221 | if (r < 0) | |
dd2bc473 | 1222 | goto fail_unlock; |
1d3576fd SW |
1223 | goto retry_locked; |
1224 | } | |
1225 | ||
1226 | if (PageUptodate(page)) { | |
1227 | dout(" page %p already uptodate\n", page); | |
1228 | return 0; | |
1229 | } | |
1230 | ||
1231 | /* full page? */ | |
09cbfeaf | 1232 | if (pos_in_page == 0 && len == PAGE_SIZE) |
1d3576fd SW |
1233 | return 0; |
1234 | ||
1235 | /* past end of file? */ | |
99c88e69 | 1236 | i_size = i_size_read(inode); |
1d3576fd | 1237 | |
1d3576fd SW |
1238 | if (page_off >= i_size || |
1239 | (pos_in_page == 0 && (pos+len) >= i_size && | |
09cbfeaf | 1240 | end_in_page - pos_in_page != PAGE_SIZE)) { |
1d3576fd | 1241 | dout(" zeroing %p 0 - %d and %d - %d\n", |
09cbfeaf | 1242 | page, pos_in_page, end_in_page, (int)PAGE_SIZE); |
1d3576fd SW |
1243 | zero_user_segments(page, |
1244 | 0, pos_in_page, | |
09cbfeaf | 1245 | end_in_page, PAGE_SIZE); |
1d3576fd SW |
1246 | return 0; |
1247 | } | |
1248 | ||
1249 | /* we need to read it. */ | |
dd2bc473 YZ |
1250 | r = ceph_do_readpage(file, page); |
1251 | if (r < 0) { | |
1252 | if (r == -EINPROGRESS) | |
1253 | return -EAGAIN; | |
1254 | goto fail_unlock; | |
1255 | } | |
1d3576fd | 1256 | goto retry_locked; |
dd2bc473 | 1257 | fail_unlock: |
1d3576fd SW |
1258 | unlock_page(page); |
1259 | return r; | |
1260 | } | |
1261 | ||
4af6b225 YS |
1262 | /* |
1263 | * We are only allowed to write into/dirty the page if the page is | |
1264 | * clean, or already dirty within the same snap context. | |
1265 | */ | |
1266 | static int ceph_write_begin(struct file *file, struct address_space *mapping, | |
1267 | loff_t pos, unsigned len, unsigned flags, | |
1268 | struct page **pagep, void **fsdata) | |
1269 | { | |
496ad9aa | 1270 | struct inode *inode = file_inode(file); |
4af6b225 | 1271 | struct page *page; |
09cbfeaf | 1272 | pgoff_t index = pos >> PAGE_SHIFT; |
7971bd92 | 1273 | int r; |
4af6b225 YS |
1274 | |
1275 | do { | |
8f883c24 | 1276 | /* get a page */ |
4af6b225 | 1277 | page = grab_cache_page_write_begin(mapping, index, 0); |
7971bd92 SW |
1278 | if (!page) |
1279 | return -ENOMEM; | |
4af6b225 YS |
1280 | |
1281 | dout("write_begin file %p inode %p page %p %d~%d\n", file, | |
213c99ee | 1282 | inode, page, (int)pos, (int)len); |
4af6b225 YS |
1283 | |
1284 | r = ceph_update_writeable_page(file, pos, len, page); | |
c1d00b2d | 1285 | if (r < 0) |
09cbfeaf | 1286 | put_page(page); |
c1d00b2d TK |
1287 | else |
1288 | *pagep = page; | |
4af6b225 YS |
1289 | } while (r == -EAGAIN); |
1290 | ||
1291 | return r; | |
1292 | } | |
1293 | ||
1d3576fd SW |
1294 | /* |
1295 | * we don't do anything in here that simple_write_end doesn't do | |
5dda377c | 1296 | * except adjust dirty page accounting |
1d3576fd SW |
1297 | */ |
1298 | static int ceph_write_end(struct file *file, struct address_space *mapping, | |
1299 | loff_t pos, unsigned len, unsigned copied, | |
1300 | struct page *page, void *fsdata) | |
1301 | { | |
496ad9aa | 1302 | struct inode *inode = file_inode(file); |
efb0ca76 | 1303 | bool check_cap = false; |
1d3576fd SW |
1304 | |
1305 | dout("write_end file %p inode %p page %p %d~%d (%d)\n", file, | |
1306 | inode, page, (int)pos, (int)copied, (int)len); | |
1307 | ||
1308 | /* zero the stale part of the page if we did a short copy */ | |
b9de313c AV |
1309 | if (!PageUptodate(page)) { |
1310 | if (copied < len) { | |
1311 | copied = 0; | |
1312 | goto out; | |
1313 | } | |
1314 | SetPageUptodate(page); | |
1315 | } | |
1d3576fd SW |
1316 | |
1317 | /* did file size increase? */ | |
99c88e69 | 1318 | if (pos+copied > i_size_read(inode)) |
1d3576fd SW |
1319 | check_cap = ceph_inode_set_size(inode, pos+copied); |
1320 | ||
1d3576fd SW |
1321 | set_page_dirty(page); |
1322 | ||
b9de313c | 1323 | out: |
1d3576fd | 1324 | unlock_page(page); |
09cbfeaf | 1325 | put_page(page); |
1d3576fd SW |
1326 | |
1327 | if (check_cap) | |
1328 | ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL); | |
1329 | ||
1330 | return copied; | |
1331 | } | |
1332 | ||
1333 | /* | |
1334 | * we set .direct_IO to indicate direct io is supported, but since we | |
1335 | * intercept O_DIRECT reads and writes early, this function should | |
1336 | * never get called. | |
1337 | */ | |
c8b8e32d | 1338 | static ssize_t ceph_direct_io(struct kiocb *iocb, struct iov_iter *iter) |
1d3576fd SW |
1339 | { |
1340 | WARN_ON(1); | |
1341 | return -EINVAL; | |
1342 | } | |
1343 | ||
1344 | const struct address_space_operations ceph_aops = { | |
1345 | .readpage = ceph_readpage, | |
1346 | .readpages = ceph_readpages, | |
1347 | .writepage = ceph_writepage, | |
1348 | .writepages = ceph_writepages_start, | |
1349 | .write_begin = ceph_write_begin, | |
1350 | .write_end = ceph_write_end, | |
1351 | .set_page_dirty = ceph_set_page_dirty, | |
1352 | .invalidatepage = ceph_invalidatepage, | |
1353 | .releasepage = ceph_releasepage, | |
1354 | .direct_IO = ceph_direct_io, | |
1355 | }; | |
1356 | ||
4f7e89f6 YZ |
1357 | static void ceph_block_sigs(sigset_t *oldset) |
1358 | { | |
1359 | sigset_t mask; | |
1360 | siginitsetinv(&mask, sigmask(SIGKILL)); | |
1361 | sigprocmask(SIG_BLOCK, &mask, oldset); | |
1362 | } | |
1363 | ||
1364 | static void ceph_restore_sigs(sigset_t *oldset) | |
1365 | { | |
1366 | sigprocmask(SIG_SETMASK, oldset, NULL); | |
1367 | } | |
1d3576fd SW |
1368 | |
1369 | /* | |
1370 | * vm ops | |
1371 | */ | |
11bac800 | 1372 | static int ceph_filemap_fault(struct vm_fault *vmf) |
61f68816 | 1373 | { |
11bac800 | 1374 | struct vm_area_struct *vma = vmf->vma; |
61f68816 YZ |
1375 | struct inode *inode = file_inode(vma->vm_file); |
1376 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1377 | struct ceph_file_info *fi = vma->vm_file->private_data; | |
3738daa6 | 1378 | struct page *pinned_page = NULL; |
09cbfeaf | 1379 | loff_t off = vmf->pgoff << PAGE_SHIFT; |
61f68816 | 1380 | int want, got, ret; |
4f7e89f6 YZ |
1381 | sigset_t oldset; |
1382 | ||
1383 | ceph_block_sigs(&oldset); | |
61f68816 YZ |
1384 | |
1385 | dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n", | |
09cbfeaf | 1386 | inode, ceph_vinop(inode), off, (size_t)PAGE_SIZE); |
61f68816 YZ |
1387 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
1388 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
1389 | else | |
1390 | want = CEPH_CAP_FILE_CACHE; | |
4f7e89f6 YZ |
1391 | |
1392 | got = 0; | |
1393 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page); | |
6ce026e4 | 1394 | if (ret < 0) |
4f7e89f6 | 1395 | goto out_restore; |
6ce026e4 | 1396 | |
61f68816 | 1397 | dout("filemap_fault %p %llu~%zd got cap refs on %s\n", |
09cbfeaf | 1398 | inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got)); |
61f68816 | 1399 | |
83701246 | 1400 | if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) || |
2b1ac852 YZ |
1401 | ci->i_inline_version == CEPH_INLINE_NONE) { |
1402 | current->journal_info = vma->vm_file; | |
11bac800 | 1403 | ret = filemap_fault(vmf); |
2b1ac852 YZ |
1404 | current->journal_info = NULL; |
1405 | } else | |
83701246 | 1406 | ret = -EAGAIN; |
61f68816 YZ |
1407 | |
1408 | dout("filemap_fault %p %llu~%zd dropping cap refs on %s ret %d\n", | |
09cbfeaf | 1409 | inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got), ret); |
3738daa6 | 1410 | if (pinned_page) |
09cbfeaf | 1411 | put_page(pinned_page); |
61f68816 YZ |
1412 | ceph_put_cap_refs(ci, got); |
1413 | ||
83701246 | 1414 | if (ret != -EAGAIN) |
4f7e89f6 | 1415 | goto out_restore; |
83701246 YZ |
1416 | |
1417 | /* read inline data */ | |
09cbfeaf | 1418 | if (off >= PAGE_SIZE) { |
83701246 YZ |
1419 | /* does not support inline data > PAGE_SIZE */ |
1420 | ret = VM_FAULT_SIGBUS; | |
1421 | } else { | |
1422 | int ret1; | |
1423 | struct address_space *mapping = inode->i_mapping; | |
1424 | struct page *page = find_or_create_page(mapping, 0, | |
c62d2555 MH |
1425 | mapping_gfp_constraint(mapping, |
1426 | ~__GFP_FS)); | |
83701246 YZ |
1427 | if (!page) { |
1428 | ret = VM_FAULT_OOM; | |
4f7e89f6 | 1429 | goto out_inline; |
83701246 YZ |
1430 | } |
1431 | ret1 = __ceph_do_getattr(inode, page, | |
1432 | CEPH_STAT_CAP_INLINE_DATA, true); | |
1433 | if (ret1 < 0 || off >= i_size_read(inode)) { | |
1434 | unlock_page(page); | |
09cbfeaf | 1435 | put_page(page); |
6ce026e4 YZ |
1436 | if (ret1 < 0) |
1437 | ret = ret1; | |
1438 | else | |
1439 | ret = VM_FAULT_SIGBUS; | |
4f7e89f6 | 1440 | goto out_inline; |
83701246 | 1441 | } |
09cbfeaf KS |
1442 | if (ret1 < PAGE_SIZE) |
1443 | zero_user_segment(page, ret1, PAGE_SIZE); | |
83701246 YZ |
1444 | else |
1445 | flush_dcache_page(page); | |
1446 | SetPageUptodate(page); | |
1447 | vmf->page = page; | |
1448 | ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED; | |
4f7e89f6 YZ |
1449 | out_inline: |
1450 | dout("filemap_fault %p %llu~%zd read inline data ret %d\n", | |
1451 | inode, off, (size_t)PAGE_SIZE, ret); | |
83701246 | 1452 | } |
4f7e89f6 YZ |
1453 | out_restore: |
1454 | ceph_restore_sigs(&oldset); | |
6ce026e4 YZ |
1455 | if (ret < 0) |
1456 | ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS; | |
1457 | ||
61f68816 YZ |
1458 | return ret; |
1459 | } | |
1d3576fd SW |
1460 | |
1461 | /* | |
1462 | * Reuse write_begin here for simplicity. | |
1463 | */ | |
11bac800 | 1464 | static int ceph_page_mkwrite(struct vm_fault *vmf) |
1d3576fd | 1465 | { |
11bac800 | 1466 | struct vm_area_struct *vma = vmf->vma; |
496ad9aa | 1467 | struct inode *inode = file_inode(vma->vm_file); |
61f68816 YZ |
1468 | struct ceph_inode_info *ci = ceph_inode(inode); |
1469 | struct ceph_file_info *fi = vma->vm_file->private_data; | |
f66fd9f0 | 1470 | struct ceph_cap_flush *prealloc_cf; |
61f68816 | 1471 | struct page *page = vmf->page; |
6285bc23 | 1472 | loff_t off = page_offset(page); |
61f68816 YZ |
1473 | loff_t size = i_size_read(inode); |
1474 | size_t len; | |
1475 | int want, got, ret; | |
4f7e89f6 | 1476 | sigset_t oldset; |
3ca9c3bd | 1477 | |
f66fd9f0 YZ |
1478 | prealloc_cf = ceph_alloc_cap_flush(); |
1479 | if (!prealloc_cf) | |
6ce026e4 | 1480 | return VM_FAULT_OOM; |
f66fd9f0 | 1481 | |
4f7e89f6 | 1482 | ceph_block_sigs(&oldset); |
f66fd9f0 | 1483 | |
28127bdd YZ |
1484 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
1485 | struct page *locked_page = NULL; | |
1486 | if (off == 0) { | |
1487 | lock_page(page); | |
1488 | locked_page = page; | |
1489 | } | |
1490 | ret = ceph_uninline_data(vma->vm_file, locked_page); | |
1491 | if (locked_page) | |
1492 | unlock_page(locked_page); | |
6ce026e4 | 1493 | if (ret < 0) |
f66fd9f0 | 1494 | goto out_free; |
28127bdd YZ |
1495 | } |
1496 | ||
09cbfeaf KS |
1497 | if (off + PAGE_SIZE <= size) |
1498 | len = PAGE_SIZE; | |
1d3576fd | 1499 | else |
09cbfeaf | 1500 | len = size & ~PAGE_MASK; |
1d3576fd | 1501 | |
61f68816 YZ |
1502 | dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n", |
1503 | inode, ceph_vinop(inode), off, len, size); | |
1504 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1505 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1506 | else | |
1507 | want = CEPH_CAP_FILE_BUFFER; | |
4f7e89f6 YZ |
1508 | |
1509 | got = 0; | |
1510 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, off + len, | |
1511 | &got, NULL); | |
6ce026e4 | 1512 | if (ret < 0) |
4f7e89f6 | 1513 | goto out_free; |
6ce026e4 | 1514 | |
61f68816 YZ |
1515 | dout("page_mkwrite %p %llu~%zd got cap refs on %s\n", |
1516 | inode, off, len, ceph_cap_string(got)); | |
1517 | ||
1518 | /* Update time before taking page lock */ | |
1519 | file_update_time(vma->vm_file); | |
4af6b225 | 1520 | |
f0b33df5 YZ |
1521 | do { |
1522 | lock_page(page); | |
4af6b225 | 1523 | |
f0b33df5 YZ |
1524 | if ((off > size) || (page->mapping != inode->i_mapping)) { |
1525 | unlock_page(page); | |
1526 | ret = VM_FAULT_NOPAGE; | |
1527 | break; | |
1528 | } | |
1529 | ||
1530 | ret = ceph_update_writeable_page(vma->vm_file, off, len, page); | |
1531 | if (ret >= 0) { | |
1532 | /* success. we'll keep the page locked. */ | |
1533 | set_page_dirty(page); | |
1534 | ret = VM_FAULT_LOCKED; | |
1535 | } | |
1536 | } while (ret == -EAGAIN); | |
4af6b225 | 1537 | |
28127bdd YZ |
1538 | if (ret == VM_FAULT_LOCKED || |
1539 | ci->i_inline_version != CEPH_INLINE_NONE) { | |
61f68816 YZ |
1540 | int dirty; |
1541 | spin_lock(&ci->i_ceph_lock); | |
28127bdd | 1542 | ci->i_inline_version = CEPH_INLINE_NONE; |
f66fd9f0 YZ |
1543 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
1544 | &prealloc_cf); | |
61f68816 YZ |
1545 | spin_unlock(&ci->i_ceph_lock); |
1546 | if (dirty) | |
1547 | __mark_inode_dirty(inode, dirty); | |
1548 | } | |
1549 | ||
1550 | dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %d\n", | |
1551 | inode, off, len, ceph_cap_string(got), ret); | |
1552 | ceph_put_cap_refs(ci, got); | |
f66fd9f0 | 1553 | out_free: |
4f7e89f6 | 1554 | ceph_restore_sigs(&oldset); |
f66fd9f0 | 1555 | ceph_free_cap_flush(prealloc_cf); |
6ce026e4 YZ |
1556 | if (ret < 0) |
1557 | ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS; | |
1d3576fd SW |
1558 | return ret; |
1559 | } | |
1560 | ||
31c542a1 YZ |
1561 | void ceph_fill_inline_data(struct inode *inode, struct page *locked_page, |
1562 | char *data, size_t len) | |
1563 | { | |
1564 | struct address_space *mapping = inode->i_mapping; | |
1565 | struct page *page; | |
1566 | ||
1567 | if (locked_page) { | |
1568 | page = locked_page; | |
1569 | } else { | |
1570 | if (i_size_read(inode) == 0) | |
1571 | return; | |
1572 | page = find_or_create_page(mapping, 0, | |
c62d2555 MH |
1573 | mapping_gfp_constraint(mapping, |
1574 | ~__GFP_FS)); | |
31c542a1 YZ |
1575 | if (!page) |
1576 | return; | |
1577 | if (PageUptodate(page)) { | |
1578 | unlock_page(page); | |
09cbfeaf | 1579 | put_page(page); |
31c542a1 YZ |
1580 | return; |
1581 | } | |
1582 | } | |
1583 | ||
0668ff52 | 1584 | dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n", |
31c542a1 YZ |
1585 | inode, ceph_vinop(inode), len, locked_page); |
1586 | ||
1587 | if (len > 0) { | |
1588 | void *kaddr = kmap_atomic(page); | |
1589 | memcpy(kaddr, data, len); | |
1590 | kunmap_atomic(kaddr); | |
1591 | } | |
1592 | ||
1593 | if (page != locked_page) { | |
09cbfeaf KS |
1594 | if (len < PAGE_SIZE) |
1595 | zero_user_segment(page, len, PAGE_SIZE); | |
31c542a1 YZ |
1596 | else |
1597 | flush_dcache_page(page); | |
1598 | ||
1599 | SetPageUptodate(page); | |
1600 | unlock_page(page); | |
09cbfeaf | 1601 | put_page(page); |
31c542a1 YZ |
1602 | } |
1603 | } | |
1604 | ||
28127bdd YZ |
1605 | int ceph_uninline_data(struct file *filp, struct page *locked_page) |
1606 | { | |
1607 | struct inode *inode = file_inode(filp); | |
1608 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1609 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
1610 | struct ceph_osd_request *req; | |
1611 | struct page *page = NULL; | |
1612 | u64 len, inline_version; | |
1613 | int err = 0; | |
1614 | bool from_pagecache = false; | |
1615 | ||
1616 | spin_lock(&ci->i_ceph_lock); | |
1617 | inline_version = ci->i_inline_version; | |
1618 | spin_unlock(&ci->i_ceph_lock); | |
1619 | ||
1620 | dout("uninline_data %p %llx.%llx inline_version %llu\n", | |
1621 | inode, ceph_vinop(inode), inline_version); | |
1622 | ||
1623 | if (inline_version == 1 || /* initial version, no data */ | |
1624 | inline_version == CEPH_INLINE_NONE) | |
1625 | goto out; | |
1626 | ||
1627 | if (locked_page) { | |
1628 | page = locked_page; | |
1629 | WARN_ON(!PageUptodate(page)); | |
1630 | } else if (ceph_caps_issued(ci) & | |
1631 | (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) { | |
1632 | page = find_get_page(inode->i_mapping, 0); | |
1633 | if (page) { | |
1634 | if (PageUptodate(page)) { | |
1635 | from_pagecache = true; | |
1636 | lock_page(page); | |
1637 | } else { | |
09cbfeaf | 1638 | put_page(page); |
28127bdd YZ |
1639 | page = NULL; |
1640 | } | |
1641 | } | |
1642 | } | |
1643 | ||
1644 | if (page) { | |
1645 | len = i_size_read(inode); | |
09cbfeaf KS |
1646 | if (len > PAGE_SIZE) |
1647 | len = PAGE_SIZE; | |
28127bdd YZ |
1648 | } else { |
1649 | page = __page_cache_alloc(GFP_NOFS); | |
1650 | if (!page) { | |
1651 | err = -ENOMEM; | |
1652 | goto out; | |
1653 | } | |
1654 | err = __ceph_do_getattr(inode, page, | |
1655 | CEPH_STAT_CAP_INLINE_DATA, true); | |
1656 | if (err < 0) { | |
1657 | /* no inline data */ | |
1658 | if (err == -ENODATA) | |
1659 | err = 0; | |
1660 | goto out; | |
1661 | } | |
1662 | len = err; | |
1663 | } | |
1664 | ||
1665 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1666 | ceph_vino(inode), 0, &len, 0, 1, | |
54ea0046 | 1667 | CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE, |
34b759b4 | 1668 | NULL, 0, 0, false); |
28127bdd YZ |
1669 | if (IS_ERR(req)) { |
1670 | err = PTR_ERR(req); | |
1671 | goto out; | |
1672 | } | |
1673 | ||
bb873b53 | 1674 | req->r_mtime = inode->i_mtime; |
28127bdd YZ |
1675 | err = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
1676 | if (!err) | |
1677 | err = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1678 | ceph_osdc_put_request(req); | |
1679 | if (err < 0) | |
1680 | goto out; | |
1681 | ||
1682 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1683 | ceph_vino(inode), 0, &len, 1, 3, | |
54ea0046 | 1684 | CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, |
34b759b4 ID |
1685 | NULL, ci->i_truncate_seq, |
1686 | ci->i_truncate_size, false); | |
28127bdd YZ |
1687 | if (IS_ERR(req)) { |
1688 | err = PTR_ERR(req); | |
1689 | goto out; | |
1690 | } | |
1691 | ||
1692 | osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false); | |
1693 | ||
ec137c10 YZ |
1694 | { |
1695 | __le64 xattr_buf = cpu_to_le64(inline_version); | |
1696 | err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR, | |
1697 | "inline_version", &xattr_buf, | |
1698 | sizeof(xattr_buf), | |
1699 | CEPH_OSD_CMPXATTR_OP_GT, | |
1700 | CEPH_OSD_CMPXATTR_MODE_U64); | |
1701 | if (err) | |
1702 | goto out_put; | |
1703 | } | |
1704 | ||
1705 | { | |
1706 | char xattr_buf[32]; | |
1707 | int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf), | |
1708 | "%llu", inline_version); | |
1709 | err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR, | |
1710 | "inline_version", | |
1711 | xattr_buf, xattr_len, 0, 0); | |
1712 | if (err) | |
1713 | goto out_put; | |
1714 | } | |
28127bdd | 1715 | |
bb873b53 | 1716 | req->r_mtime = inode->i_mtime; |
28127bdd YZ |
1717 | err = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
1718 | if (!err) | |
1719 | err = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1720 | out_put: | |
1721 | ceph_osdc_put_request(req); | |
1722 | if (err == -ECANCELED) | |
1723 | err = 0; | |
1724 | out: | |
1725 | if (page && page != locked_page) { | |
1726 | if (from_pagecache) { | |
1727 | unlock_page(page); | |
09cbfeaf | 1728 | put_page(page); |
28127bdd YZ |
1729 | } else |
1730 | __free_pages(page, 0); | |
1731 | } | |
1732 | ||
1733 | dout("uninline_data %p %llx.%llx inline_version %llu = %d\n", | |
1734 | inode, ceph_vinop(inode), inline_version, err); | |
1735 | return err; | |
1736 | } | |
1737 | ||
7cbea8dc | 1738 | static const struct vm_operations_struct ceph_vmops = { |
61f68816 | 1739 | .fault = ceph_filemap_fault, |
1d3576fd SW |
1740 | .page_mkwrite = ceph_page_mkwrite, |
1741 | }; | |
1742 | ||
1743 | int ceph_mmap(struct file *file, struct vm_area_struct *vma) | |
1744 | { | |
1745 | struct address_space *mapping = file->f_mapping; | |
1746 | ||
1747 | if (!mapping->a_ops->readpage) | |
1748 | return -ENOEXEC; | |
1749 | file_accessed(file); | |
1750 | vma->vm_ops = &ceph_vmops; | |
1d3576fd SW |
1751 | return 0; |
1752 | } | |
10183a69 YZ |
1753 | |
1754 | enum { | |
1755 | POOL_READ = 1, | |
1756 | POOL_WRITE = 2, | |
1757 | }; | |
1758 | ||
779fe0fb YZ |
1759 | static int __ceph_pool_perm_get(struct ceph_inode_info *ci, |
1760 | s64 pool, struct ceph_string *pool_ns) | |
10183a69 YZ |
1761 | { |
1762 | struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode); | |
1763 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
1764 | struct ceph_osd_request *rd_req = NULL, *wr_req = NULL; | |
1765 | struct rb_node **p, *parent; | |
1766 | struct ceph_pool_perm *perm; | |
1767 | struct page **pages; | |
779fe0fb | 1768 | size_t pool_ns_len; |
10183a69 YZ |
1769 | int err = 0, err2 = 0, have = 0; |
1770 | ||
1771 | down_read(&mdsc->pool_perm_rwsem); | |
1772 | p = &mdsc->pool_perm_tree.rb_node; | |
1773 | while (*p) { | |
1774 | perm = rb_entry(*p, struct ceph_pool_perm, node); | |
1775 | if (pool < perm->pool) | |
1776 | p = &(*p)->rb_left; | |
1777 | else if (pool > perm->pool) | |
1778 | p = &(*p)->rb_right; | |
1779 | else { | |
779fe0fb YZ |
1780 | int ret = ceph_compare_string(pool_ns, |
1781 | perm->pool_ns, | |
1782 | perm->pool_ns_len); | |
1783 | if (ret < 0) | |
1784 | p = &(*p)->rb_left; | |
1785 | else if (ret > 0) | |
1786 | p = &(*p)->rb_right; | |
1787 | else { | |
1788 | have = perm->perm; | |
1789 | break; | |
1790 | } | |
10183a69 YZ |
1791 | } |
1792 | } | |
1793 | up_read(&mdsc->pool_perm_rwsem); | |
1794 | if (*p) | |
1795 | goto out; | |
1796 | ||
779fe0fb YZ |
1797 | if (pool_ns) |
1798 | dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n", | |
1799 | pool, (int)pool_ns->len, pool_ns->str); | |
1800 | else | |
1801 | dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool); | |
10183a69 YZ |
1802 | |
1803 | down_write(&mdsc->pool_perm_rwsem); | |
779fe0fb | 1804 | p = &mdsc->pool_perm_tree.rb_node; |
10183a69 YZ |
1805 | parent = NULL; |
1806 | while (*p) { | |
1807 | parent = *p; | |
1808 | perm = rb_entry(parent, struct ceph_pool_perm, node); | |
1809 | if (pool < perm->pool) | |
1810 | p = &(*p)->rb_left; | |
1811 | else if (pool > perm->pool) | |
1812 | p = &(*p)->rb_right; | |
1813 | else { | |
779fe0fb YZ |
1814 | int ret = ceph_compare_string(pool_ns, |
1815 | perm->pool_ns, | |
1816 | perm->pool_ns_len); | |
1817 | if (ret < 0) | |
1818 | p = &(*p)->rb_left; | |
1819 | else if (ret > 0) | |
1820 | p = &(*p)->rb_right; | |
1821 | else { | |
1822 | have = perm->perm; | |
1823 | break; | |
1824 | } | |
10183a69 YZ |
1825 | } |
1826 | } | |
1827 | if (*p) { | |
1828 | up_write(&mdsc->pool_perm_rwsem); | |
1829 | goto out; | |
1830 | } | |
1831 | ||
34b759b4 | 1832 | rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL, |
10183a69 YZ |
1833 | 1, false, GFP_NOFS); |
1834 | if (!rd_req) { | |
1835 | err = -ENOMEM; | |
1836 | goto out_unlock; | |
1837 | } | |
1838 | ||
1839 | rd_req->r_flags = CEPH_OSD_FLAG_READ; | |
1840 | osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0); | |
1841 | rd_req->r_base_oloc.pool = pool; | |
779fe0fb YZ |
1842 | if (pool_ns) |
1843 | rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns); | |
d30291b9 | 1844 | ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino); |
10183a69 | 1845 | |
13d1ad16 ID |
1846 | err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS); |
1847 | if (err) | |
1848 | goto out_unlock; | |
10183a69 | 1849 | |
34b759b4 | 1850 | wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL, |
10183a69 YZ |
1851 | 1, false, GFP_NOFS); |
1852 | if (!wr_req) { | |
1853 | err = -ENOMEM; | |
1854 | goto out_unlock; | |
1855 | } | |
1856 | ||
54ea0046 | 1857 | wr_req->r_flags = CEPH_OSD_FLAG_WRITE; |
10183a69 | 1858 | osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL); |
63244fa1 | 1859 | ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc); |
d30291b9 | 1860 | ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid); |
10183a69 | 1861 | |
13d1ad16 ID |
1862 | err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS); |
1863 | if (err) | |
1864 | goto out_unlock; | |
10183a69 YZ |
1865 | |
1866 | /* one page should be large enough for STAT data */ | |
1867 | pages = ceph_alloc_page_vector(1, GFP_KERNEL); | |
1868 | if (IS_ERR(pages)) { | |
1869 | err = PTR_ERR(pages); | |
1870 | goto out_unlock; | |
1871 | } | |
1872 | ||
1873 | osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE, | |
1874 | 0, false, true); | |
10183a69 YZ |
1875 | err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false); |
1876 | ||
bb873b53 | 1877 | wr_req->r_mtime = ci->vfs_inode.i_mtime; |
a1f4020a | 1878 | wr_req->r_abort_on_full = true; |
10183a69 YZ |
1879 | err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false); |
1880 | ||
1881 | if (!err) | |
1882 | err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req); | |
1883 | if (!err2) | |
1884 | err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req); | |
1885 | ||
1886 | if (err >= 0 || err == -ENOENT) | |
1887 | have |= POOL_READ; | |
1888 | else if (err != -EPERM) | |
1889 | goto out_unlock; | |
1890 | ||
1891 | if (err2 == 0 || err2 == -EEXIST) | |
1892 | have |= POOL_WRITE; | |
1893 | else if (err2 != -EPERM) { | |
1894 | err = err2; | |
1895 | goto out_unlock; | |
1896 | } | |
1897 | ||
779fe0fb YZ |
1898 | pool_ns_len = pool_ns ? pool_ns->len : 0; |
1899 | perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS); | |
10183a69 YZ |
1900 | if (!perm) { |
1901 | err = -ENOMEM; | |
1902 | goto out_unlock; | |
1903 | } | |
1904 | ||
1905 | perm->pool = pool; | |
1906 | perm->perm = have; | |
779fe0fb YZ |
1907 | perm->pool_ns_len = pool_ns_len; |
1908 | if (pool_ns_len > 0) | |
1909 | memcpy(perm->pool_ns, pool_ns->str, pool_ns_len); | |
1910 | perm->pool_ns[pool_ns_len] = 0; | |
1911 | ||
10183a69 YZ |
1912 | rb_link_node(&perm->node, parent, p); |
1913 | rb_insert_color(&perm->node, &mdsc->pool_perm_tree); | |
1914 | err = 0; | |
1915 | out_unlock: | |
1916 | up_write(&mdsc->pool_perm_rwsem); | |
1917 | ||
3ed97d63 ID |
1918 | ceph_osdc_put_request(rd_req); |
1919 | ceph_osdc_put_request(wr_req); | |
10183a69 YZ |
1920 | out: |
1921 | if (!err) | |
1922 | err = have; | |
779fe0fb YZ |
1923 | if (pool_ns) |
1924 | dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n", | |
1925 | pool, (int)pool_ns->len, pool_ns->str, err); | |
1926 | else | |
1927 | dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err); | |
10183a69 YZ |
1928 | return err; |
1929 | } | |
1930 | ||
1931 | int ceph_pool_perm_check(struct ceph_inode_info *ci, int need) | |
1932 | { | |
7627151e | 1933 | s64 pool; |
779fe0fb | 1934 | struct ceph_string *pool_ns; |
10183a69 YZ |
1935 | int ret, flags; |
1936 | ||
80e80fbb YZ |
1937 | if (ci->i_vino.snap != CEPH_NOSNAP) { |
1938 | /* | |
1939 | * Pool permission check needs to write to the first object. | |
1940 | * But for snapshot, head of the first object may have alread | |
1941 | * been deleted. Skip check to avoid creating orphan object. | |
1942 | */ | |
1943 | return 0; | |
1944 | } | |
1945 | ||
10183a69 YZ |
1946 | if (ceph_test_mount_opt(ceph_inode_to_client(&ci->vfs_inode), |
1947 | NOPOOLPERM)) | |
1948 | return 0; | |
1949 | ||
1950 | spin_lock(&ci->i_ceph_lock); | |
1951 | flags = ci->i_ceph_flags; | |
7627151e | 1952 | pool = ci->i_layout.pool_id; |
10183a69 YZ |
1953 | spin_unlock(&ci->i_ceph_lock); |
1954 | check: | |
1955 | if (flags & CEPH_I_POOL_PERM) { | |
1956 | if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) { | |
7627151e | 1957 | dout("ceph_pool_perm_check pool %lld no read perm\n", |
10183a69 YZ |
1958 | pool); |
1959 | return -EPERM; | |
1960 | } | |
1961 | if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) { | |
7627151e | 1962 | dout("ceph_pool_perm_check pool %lld no write perm\n", |
10183a69 YZ |
1963 | pool); |
1964 | return -EPERM; | |
1965 | } | |
1966 | return 0; | |
1967 | } | |
1968 | ||
779fe0fb YZ |
1969 | pool_ns = ceph_try_get_string(ci->i_layout.pool_ns); |
1970 | ret = __ceph_pool_perm_get(ci, pool, pool_ns); | |
1971 | ceph_put_string(pool_ns); | |
10183a69 YZ |
1972 | if (ret < 0) |
1973 | return ret; | |
1974 | ||
1975 | flags = CEPH_I_POOL_PERM; | |
1976 | if (ret & POOL_READ) | |
1977 | flags |= CEPH_I_POOL_RD; | |
1978 | if (ret & POOL_WRITE) | |
1979 | flags |= CEPH_I_POOL_WR; | |
1980 | ||
1981 | spin_lock(&ci->i_ceph_lock); | |
779fe0fb YZ |
1982 | if (pool == ci->i_layout.pool_id && |
1983 | pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) { | |
1984 | ci->i_ceph_flags |= flags; | |
10183a69 | 1985 | } else { |
7627151e | 1986 | pool = ci->i_layout.pool_id; |
10183a69 YZ |
1987 | flags = ci->i_ceph_flags; |
1988 | } | |
1989 | spin_unlock(&ci->i_ceph_lock); | |
1990 | goto check; | |
1991 | } | |
1992 | ||
1993 | void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc) | |
1994 | { | |
1995 | struct ceph_pool_perm *perm; | |
1996 | struct rb_node *n; | |
1997 | ||
1998 | while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) { | |
1999 | n = rb_first(&mdsc->pool_perm_tree); | |
2000 | perm = rb_entry(n, struct ceph_pool_perm, node); | |
2001 | rb_erase(n, &mdsc->pool_perm_tree); | |
2002 | kfree(perm); | |
2003 | } | |
2004 | } |