Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
3d14c5d2 | 2 | #include <linux/ceph/ceph_debug.h> |
1d3576fd SW |
3 | |
4 | #include <linux/backing-dev.h> | |
5 | #include <linux/fs.h> | |
6 | #include <linux/mm.h> | |
d7bdba1c | 7 | #include <linux/swap.h> |
1d3576fd | 8 | #include <linux/pagemap.h> |
5a0e3ad6 | 9 | #include <linux/slab.h> |
1d3576fd SW |
10 | #include <linux/pagevec.h> |
11 | #include <linux/task_io_accounting_ops.h> | |
f361bf4a | 12 | #include <linux/signal.h> |
5c308356 | 13 | #include <linux/iversion.h> |
97e27aaa | 14 | #include <linux/ktime.h> |
f0702876 | 15 | #include <linux/netfs.h> |
1d3576fd SW |
16 | |
17 | #include "super.h" | |
3d14c5d2 | 18 | #include "mds_client.h" |
99ccbd22 | 19 | #include "cache.h" |
97e27aaa | 20 | #include "metric.h" |
3d14c5d2 | 21 | #include <linux/ceph/osd_client.h> |
08c1ac50 | 22 | #include <linux/ceph/striper.h> |
1d3576fd SW |
23 | |
24 | /* | |
25 | * Ceph address space ops. | |
26 | * | |
27 | * There are a few funny things going on here. | |
28 | * | |
29 | * The page->private field is used to reference a struct | |
30 | * ceph_snap_context for _every_ dirty page. This indicates which | |
31 | * snapshot the page was logically dirtied in, and thus which snap | |
32 | * context needs to be associated with the osd write during writeback. | |
33 | * | |
34 | * Similarly, struct ceph_inode_info maintains a set of counters to | |
25985edc | 35 | * count dirty pages on the inode. In the absence of snapshots, |
1d3576fd SW |
36 | * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count. |
37 | * | |
38 | * When a snapshot is taken (that is, when the client receives | |
39 | * notification that a snapshot was taken), each inode with caps and | |
40 | * with dirty pages (dirty pages implies there is a cap) gets a new | |
41 | * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending | |
42 | * order, new snaps go to the tail). The i_wrbuffer_ref_head count is | |
43 | * moved to capsnap->dirty. (Unless a sync write is currently in | |
44 | * progress. In that case, the capsnap is said to be "pending", new | |
45 | * writes cannot start, and the capsnap isn't "finalized" until the | |
46 | * write completes (or fails) and a final size/mtime for the inode for | |
47 | * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0. | |
48 | * | |
49 | * On writeback, we must submit writes to the osd IN SNAP ORDER. So, | |
50 | * we look for the first capsnap in i_cap_snaps and write out pages in | |
51 | * that snap context _only_. Then we move on to the next capsnap, | |
52 | * eventually reaching the "live" or "head" context (i.e., pages that | |
53 | * are not yet snapped) and are writing the most recently dirtied | |
54 | * pages. | |
55 | * | |
56 | * Invalidate and so forth must take care to ensure the dirty page | |
57 | * accounting is preserved. | |
58 | */ | |
59 | ||
2baba250 YS |
60 | #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10)) |
61 | #define CONGESTION_OFF_THRESH(congestion_kb) \ | |
62 | (CONGESTION_ON_THRESH(congestion_kb) - \ | |
63 | (CONGESTION_ON_THRESH(congestion_kb) >> 2)) | |
64 | ||
d801327d | 65 | static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len, |
78525c74 | 66 | struct folio *folio, void **_fsdata); |
d801327d | 67 | |
61600ef8 YZ |
68 | static inline struct ceph_snap_context *page_snap_context(struct page *page) |
69 | { | |
70 | if (PagePrivate(page)) | |
71 | return (void *)page->private; | |
72 | return NULL; | |
73 | } | |
1d3576fd SW |
74 | |
75 | /* | |
76 | * Dirty a page. Optimistically adjust accounting, on the assumption | |
77 | * that we won't race with invalidate. If we do, readjust. | |
78 | */ | |
8fb72b4a | 79 | static bool ceph_dirty_folio(struct address_space *mapping, struct folio *folio) |
1d3576fd | 80 | { |
1d3576fd SW |
81 | struct inode *inode; |
82 | struct ceph_inode_info *ci; | |
1d3576fd | 83 | struct ceph_snap_context *snapc; |
1d3576fd | 84 | |
8fb72b4a MWO |
85 | if (folio_test_dirty(folio)) { |
86 | dout("%p dirty_folio %p idx %lu -- already dirty\n", | |
87 | mapping->host, folio, folio->index); | |
642d51fb | 88 | VM_BUG_ON_FOLIO(!folio_test_private(folio), folio); |
8fb72b4a | 89 | return false; |
1d3576fd SW |
90 | } |
91 | ||
92 | inode = mapping->host; | |
93 | ci = ceph_inode(inode); | |
94 | ||
1d3576fd | 95 | /* dirty the head */ |
be655596 | 96 | spin_lock(&ci->i_ceph_lock); |
5dda377c YZ |
97 | BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference |
98 | if (__ceph_have_pending_cap_snap(ci)) { | |
99 | struct ceph_cap_snap *capsnap = | |
100 | list_last_entry(&ci->i_cap_snaps, | |
101 | struct ceph_cap_snap, | |
102 | ci_item); | |
103 | snapc = ceph_get_snap_context(capsnap->context); | |
104 | capsnap->dirty_pages++; | |
105 | } else { | |
106 | BUG_ON(!ci->i_head_snapc); | |
107 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
108 | ++ci->i_wrbuffer_ref_head; | |
109 | } | |
1d3576fd | 110 | if (ci->i_wrbuffer_ref == 0) |
0444d76a | 111 | ihold(inode); |
1d3576fd | 112 | ++ci->i_wrbuffer_ref; |
8fb72b4a | 113 | dout("%p dirty_folio %p idx %lu head %d/%d -> %d/%d " |
1d3576fd | 114 | "snapc %p seq %lld (%d snaps)\n", |
8fb72b4a | 115 | mapping->host, folio, folio->index, |
1d3576fd SW |
116 | ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1, |
117 | ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, | |
118 | snapc, snapc->seq, snapc->num_snaps); | |
be655596 | 119 | spin_unlock(&ci->i_ceph_lock); |
1d3576fd | 120 | |
7d6e1f54 | 121 | /* |
8fb72b4a | 122 | * Reference snap context in folio->private. Also set |
9872f4de | 123 | * PagePrivate so that we get invalidate_folio callback. |
7d6e1f54 | 124 | */ |
642d51fb | 125 | VM_BUG_ON_FOLIO(folio_test_private(folio), folio); |
8fb72b4a | 126 | folio_attach_private(folio, snapc); |
1d3576fd | 127 | |
8fb72b4a | 128 | return ceph_fscache_dirty_folio(mapping, folio); |
1d3576fd SW |
129 | } |
130 | ||
131 | /* | |
9872f4de MWO |
132 | * If we are truncating the full folio (i.e. offset == 0), adjust the |
133 | * dirty folio counters appropriately. Only called if there is private | |
134 | * data on the folio. | |
1d3576fd | 135 | */ |
9872f4de MWO |
136 | static void ceph_invalidate_folio(struct folio *folio, size_t offset, |
137 | size_t length) | |
1d3576fd | 138 | { |
4ce1e9ad | 139 | struct inode *inode; |
1d3576fd | 140 | struct ceph_inode_info *ci; |
379fc7fa | 141 | struct ceph_snap_context *snapc; |
1d3576fd | 142 | |
9872f4de | 143 | inode = folio->mapping->host; |
b150f5c1 MT |
144 | ci = ceph_inode(inode); |
145 | ||
9872f4de MWO |
146 | if (offset != 0 || length != folio_size(folio)) { |
147 | dout("%p invalidate_folio idx %lu partial dirty page %zu~%zu\n", | |
148 | inode, folio->index, offset, length); | |
b150f5c1 MT |
149 | return; |
150 | } | |
4ce1e9ad | 151 | |
9872f4de | 152 | WARN_ON(!folio_test_locked(folio)); |
642d51fb | 153 | if (folio_test_private(folio)) { |
9872f4de MWO |
154 | dout("%p invalidate_folio idx %lu full dirty page\n", |
155 | inode, folio->index); | |
99ccbd22 | 156 | |
9872f4de | 157 | snapc = folio_detach_private(folio); |
400e1286 JL |
158 | ceph_put_wrbuffer_cap_refs(ci, 1, snapc); |
159 | ceph_put_snap_context(snapc); | |
160 | } | |
b150f5c1 | 161 | |
9872f4de | 162 | folio_wait_fscache(folio); |
1d3576fd SW |
163 | } |
164 | ||
5e414655 | 165 | static bool ceph_release_folio(struct folio *folio, gfp_t gfp) |
1d3576fd | 166 | { |
5e414655 | 167 | struct inode *inode = folio->mapping->host; |
400e1286 | 168 | |
5e414655 MWO |
169 | dout("%llx:%llx release_folio idx %lu (%sdirty)\n", |
170 | ceph_vinop(inode), | |
171 | folio->index, folio_test_dirty(folio) ? "" : "not "); | |
400e1286 | 172 | |
5e414655 MWO |
173 | if (folio_test_private(folio)) |
174 | return false; | |
99ccbd22 | 175 | |
5e414655 | 176 | if (folio_test_fscache(folio)) { |
d7bdba1c | 177 | if (current_is_kswapd() || !(gfp & __GFP_FS)) |
5e414655 MWO |
178 | return false; |
179 | folio_wait_fscache(folio); | |
7c46b318 | 180 | } |
400e1286 | 181 | ceph_fscache_note_page_release(inode); |
5e414655 | 182 | return true; |
1d3576fd SW |
183 | } |
184 | ||
6a19114b | 185 | static void ceph_netfs_expand_readahead(struct netfs_io_request *rreq) |
f0702876 | 186 | { |
a25cedb4 | 187 | struct inode *inode = rreq->inode; |
f0702876 JL |
188 | struct ceph_inode_info *ci = ceph_inode(inode); |
189 | struct ceph_file_layout *lo = &ci->i_layout; | |
190 | u32 blockoff; | |
191 | u64 blockno; | |
192 | ||
193 | /* Expand the start downward */ | |
194 | blockno = div_u64_rem(rreq->start, lo->stripe_unit, &blockoff); | |
195 | rreq->start = blockno * lo->stripe_unit; | |
196 | rreq->len += blockoff; | |
197 | ||
198 | /* Now, round up the length to the next block */ | |
199 | rreq->len = roundup(rreq->len, lo->stripe_unit); | |
200 | } | |
201 | ||
6a19114b | 202 | static bool ceph_netfs_clamp_length(struct netfs_io_subrequest *subreq) |
f0702876 | 203 | { |
a25cedb4 | 204 | struct inode *inode = subreq->rreq->inode; |
f0702876 JL |
205 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
206 | struct ceph_inode_info *ci = ceph_inode(inode); | |
207 | u64 objno, objoff; | |
208 | u32 xlen; | |
209 | ||
210 | /* Truncate the extent at the end of the current block */ | |
211 | ceph_calc_file_object_mapping(&ci->i_layout, subreq->start, subreq->len, | |
212 | &objno, &objoff, &xlen); | |
213 | subreq->len = min(xlen, fsc->mount_options->rsize); | |
214 | return true; | |
215 | } | |
216 | ||
217 | static void finish_netfs_read(struct ceph_osd_request *req) | |
218 | { | |
219 | struct ceph_fs_client *fsc = ceph_inode_to_client(req->r_inode); | |
220 | struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0); | |
6a19114b | 221 | struct netfs_io_subrequest *subreq = req->r_priv; |
f0702876 JL |
222 | int num_pages; |
223 | int err = req->r_result; | |
224 | ||
8ae99ae2 | 225 | ceph_update_read_metrics(&fsc->mdsc->metric, req->r_start_latency, |
903f4fec | 226 | req->r_end_latency, osd_data->length, err); |
f0702876 JL |
227 | |
228 | dout("%s: result %d subreq->len=%zu i_size=%lld\n", __func__, req->r_result, | |
229 | subreq->len, i_size_read(req->r_inode)); | |
230 | ||
231 | /* no object means success but no data */ | |
232 | if (err == -ENOENT) | |
233 | err = 0; | |
234 | else if (err == -EBLOCKLISTED) | |
235 | fsc->blocklisted = true; | |
236 | ||
237 | if (err >= 0 && err < subreq->len) | |
238 | __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags); | |
239 | ||
240 | netfs_subreq_terminated(subreq, err, true); | |
241 | ||
242 | num_pages = calc_pages_for(osd_data->alignment, osd_data->length); | |
243 | ceph_put_page_vector(osd_data->pages, num_pages, false); | |
244 | iput(req->r_inode); | |
245 | } | |
246 | ||
6a19114b | 247 | static bool ceph_netfs_issue_op_inline(struct netfs_io_subrequest *subreq) |
5b19f1eb | 248 | { |
6a19114b | 249 | struct netfs_io_request *rreq = subreq->rreq; |
5b19f1eb DH |
250 | struct inode *inode = rreq->inode; |
251 | struct ceph_mds_reply_info_parsed *rinfo; | |
252 | struct ceph_mds_reply_info_in *iinfo; | |
253 | struct ceph_mds_request *req; | |
254 | struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); | |
255 | struct ceph_inode_info *ci = ceph_inode(inode); | |
256 | struct iov_iter iter; | |
257 | ssize_t err = 0; | |
258 | size_t len; | |
5eed80fb | 259 | int mode; |
5b19f1eb DH |
260 | |
261 | __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags); | |
f18a3785 | 262 | __clear_bit(NETFS_SREQ_COPY_TO_CACHE, &subreq->flags); |
5b19f1eb DH |
263 | |
264 | if (subreq->start >= inode->i_size) | |
265 | goto out; | |
266 | ||
267 | /* We need to fetch the inline data. */ | |
5eed80fb XL |
268 | mode = ceph_try_to_choose_auth_mds(inode, CEPH_STAT_CAP_INLINE_DATA); |
269 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode); | |
5b19f1eb DH |
270 | if (IS_ERR(req)) { |
271 | err = PTR_ERR(req); | |
272 | goto out; | |
273 | } | |
274 | req->r_ino1 = ci->i_vino; | |
275 | req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INLINE_DATA); | |
276 | req->r_num_caps = 2; | |
277 | ||
278 | err = ceph_mdsc_do_request(mdsc, NULL, req); | |
279 | if (err < 0) | |
280 | goto out; | |
281 | ||
282 | rinfo = &req->r_reply_info; | |
283 | iinfo = &rinfo->targeti; | |
284 | if (iinfo->inline_version == CEPH_INLINE_NONE) { | |
285 | /* The data got uninlined */ | |
286 | ceph_mdsc_put_request(req); | |
287 | return false; | |
288 | } | |
289 | ||
290 | len = min_t(size_t, iinfo->inline_len - subreq->start, subreq->len); | |
291 | iov_iter_xarray(&iter, READ, &rreq->mapping->i_pages, subreq->start, len); | |
292 | err = copy_to_iter(iinfo->inline_data + subreq->start, len, &iter); | |
293 | if (err == 0) | |
294 | err = -EFAULT; | |
295 | ||
296 | ceph_mdsc_put_request(req); | |
297 | out: | |
298 | netfs_subreq_terminated(subreq, err, false); | |
299 | return true; | |
300 | } | |
301 | ||
f18a3785 | 302 | static void ceph_netfs_issue_read(struct netfs_io_subrequest *subreq) |
f0702876 | 303 | { |
6a19114b | 304 | struct netfs_io_request *rreq = subreq->rreq; |
a25cedb4 | 305 | struct inode *inode = rreq->inode; |
f0702876 JL |
306 | struct ceph_inode_info *ci = ceph_inode(inode); |
307 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
308 | struct ceph_osd_request *req; | |
309 | struct ceph_vino vino = ceph_vino(inode); | |
310 | struct iov_iter iter; | |
311 | struct page **pages; | |
312 | size_t page_off; | |
313 | int err = 0; | |
314 | u64 len = subreq->len; | |
315 | ||
5b19f1eb DH |
316 | if (ci->i_inline_version != CEPH_INLINE_NONE && |
317 | ceph_netfs_issue_op_inline(subreq)) | |
318 | return; | |
319 | ||
f0702876 JL |
320 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, vino, subreq->start, &len, |
321 | 0, 1, CEPH_OSD_OP_READ, | |
322 | CEPH_OSD_FLAG_READ | fsc->client->osdc.client->options->read_from_replica, | |
323 | NULL, ci->i_truncate_seq, ci->i_truncate_size, false); | |
324 | if (IS_ERR(req)) { | |
325 | err = PTR_ERR(req); | |
326 | req = NULL; | |
327 | goto out; | |
328 | } | |
329 | ||
330 | dout("%s: pos=%llu orig_len=%zu len=%llu\n", __func__, subreq->start, subreq->len, len); | |
331 | iov_iter_xarray(&iter, READ, &rreq->mapping->i_pages, subreq->start, len); | |
332 | err = iov_iter_get_pages_alloc(&iter, &pages, len, &page_off); | |
333 | if (err < 0) { | |
334 | dout("%s: iov_ter_get_pages_alloc returned %d\n", __func__, err); | |
335 | goto out; | |
336 | } | |
337 | ||
338 | /* should always give us a page-aligned read */ | |
339 | WARN_ON_ONCE(page_off); | |
340 | len = err; | |
341 | ||
342 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false); | |
343 | req->r_callback = finish_netfs_read; | |
344 | req->r_priv = subreq; | |
345 | req->r_inode = inode; | |
346 | ihold(inode); | |
347 | ||
348 | err = ceph_osdc_start_request(req->r_osdc, req, false); | |
349 | if (err) | |
350 | iput(inode); | |
351 | out: | |
352 | ceph_osdc_put_request(req); | |
353 | if (err) | |
354 | netfs_subreq_terminated(subreq, err, false); | |
355 | dout("%s: result %d\n", __func__, err); | |
356 | } | |
357 | ||
a5c9dc44 DH |
358 | static int ceph_init_request(struct netfs_io_request *rreq, struct file *file) |
359 | { | |
360 | struct inode *inode = rreq->inode; | |
361 | int got = 0, want = CEPH_CAP_FILE_CACHE; | |
362 | int ret = 0; | |
363 | ||
364 | if (rreq->origin != NETFS_READAHEAD) | |
365 | return 0; | |
366 | ||
367 | if (file) { | |
368 | struct ceph_rw_context *rw_ctx; | |
369 | struct ceph_file_info *fi = file->private_data; | |
370 | ||
371 | rw_ctx = ceph_find_rw_context(fi); | |
372 | if (rw_ctx) | |
373 | return 0; | |
374 | } | |
375 | ||
376 | /* | |
377 | * readahead callers do not necessarily hold Fcb caps | |
378 | * (e.g. fadvise, madvise). | |
379 | */ | |
380 | ret = ceph_try_get_caps(inode, CEPH_CAP_FILE_RD, want, true, &got); | |
381 | if (ret < 0) { | |
382 | dout("start_read %p, error getting cap\n", inode); | |
383 | return ret; | |
384 | } | |
385 | ||
386 | if (!(got & want)) { | |
387 | dout("start_read %p, no cache cap\n", inode); | |
388 | return -EACCES; | |
389 | } | |
390 | if (ret == 0) | |
391 | return -EACCES; | |
392 | ||
393 | rreq->netfs_priv = (void *)(uintptr_t)got; | |
394 | return 0; | |
395 | } | |
396 | ||
49870056 JL |
397 | static void ceph_readahead_cleanup(struct address_space *mapping, void *priv) |
398 | { | |
399 | struct inode *inode = mapping->host; | |
400 | struct ceph_inode_info *ci = ceph_inode(inode); | |
401 | int got = (uintptr_t)priv; | |
402 | ||
403 | if (got) | |
404 | ceph_put_cap_refs(ci, got); | |
405 | } | |
406 | ||
bc899ee1 | 407 | const struct netfs_request_ops ceph_netfs_ops = { |
a5c9dc44 | 408 | .init_request = ceph_init_request, |
f0702876 | 409 | .begin_cache_operation = ceph_begin_cache_operation, |
f18a3785 | 410 | .issue_read = ceph_netfs_issue_read, |
f0702876 JL |
411 | .expand_readahead = ceph_netfs_expand_readahead, |
412 | .clamp_length = ceph_netfs_clamp_length, | |
d801327d | 413 | .check_write_begin = ceph_netfs_check_write_begin, |
49870056 | 414 | .cleanup = ceph_readahead_cleanup, |
f0702876 JL |
415 | }; |
416 | ||
1702e797 JL |
417 | #ifdef CONFIG_CEPH_FSCACHE |
418 | static void ceph_set_page_fscache(struct page *page) | |
419 | { | |
420 | set_page_fscache(page); | |
421 | } | |
422 | ||
423 | static void ceph_fscache_write_terminated(void *priv, ssize_t error, bool was_async) | |
424 | { | |
425 | struct inode *inode = priv; | |
426 | ||
427 | if (IS_ERR_VALUE(error) && error != -ENOBUFS) | |
428 | ceph_fscache_invalidate(inode, false); | |
429 | } | |
430 | ||
431 | static void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching) | |
432 | { | |
433 | struct ceph_inode_info *ci = ceph_inode(inode); | |
434 | struct fscache_cookie *cookie = ceph_fscache_cookie(ci); | |
435 | ||
436 | fscache_write_to_cache(cookie, inode->i_mapping, off, len, i_size_read(inode), | |
437 | ceph_fscache_write_terminated, inode, caching); | |
438 | } | |
439 | #else | |
440 | static inline void ceph_set_page_fscache(struct page *page) | |
441 | { | |
442 | } | |
443 | ||
444 | static inline void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching) | |
445 | { | |
446 | } | |
447 | #endif /* CONFIG_CEPH_FSCACHE */ | |
448 | ||
1f934b00 YZ |
449 | struct ceph_writeback_ctl |
450 | { | |
451 | loff_t i_size; | |
452 | u64 truncate_size; | |
453 | u32 truncate_seq; | |
454 | bool size_stable; | |
2a2d927e | 455 | bool head_snapc; |
1f934b00 YZ |
456 | }; |
457 | ||
1d3576fd SW |
458 | /* |
459 | * Get ref for the oldest snapc for an inode with dirty data... that is, the | |
460 | * only snap context we are allowed to write back. | |
1d3576fd | 461 | */ |
1f934b00 | 462 | static struct ceph_snap_context * |
05455e11 YZ |
463 | get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl, |
464 | struct ceph_snap_context *page_snapc) | |
1d3576fd SW |
465 | { |
466 | struct ceph_inode_info *ci = ceph_inode(inode); | |
467 | struct ceph_snap_context *snapc = NULL; | |
468 | struct ceph_cap_snap *capsnap = NULL; | |
469 | ||
be655596 | 470 | spin_lock(&ci->i_ceph_lock); |
1d3576fd SW |
471 | list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { |
472 | dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap, | |
473 | capsnap->context, capsnap->dirty_pages); | |
05455e11 YZ |
474 | if (!capsnap->dirty_pages) |
475 | continue; | |
476 | ||
477 | /* get i_size, truncate_{seq,size} for page_snapc? */ | |
478 | if (snapc && capsnap->context != page_snapc) | |
479 | continue; | |
480 | ||
481 | if (ctl) { | |
482 | if (capsnap->writing) { | |
483 | ctl->i_size = i_size_read(inode); | |
484 | ctl->size_stable = false; | |
485 | } else { | |
486 | ctl->i_size = capsnap->size; | |
487 | ctl->size_stable = true; | |
1f934b00 | 488 | } |
05455e11 YZ |
489 | ctl->truncate_size = capsnap->truncate_size; |
490 | ctl->truncate_seq = capsnap->truncate_seq; | |
2a2d927e | 491 | ctl->head_snapc = false; |
1d3576fd | 492 | } |
05455e11 YZ |
493 | |
494 | if (snapc) | |
495 | break; | |
496 | ||
497 | snapc = ceph_get_snap_context(capsnap->context); | |
498 | if (!page_snapc || | |
499 | page_snapc == snapc || | |
500 | page_snapc->seq > snapc->seq) | |
501 | break; | |
1d3576fd | 502 | } |
7d8cb26d | 503 | if (!snapc && ci->i_wrbuffer_ref_head) { |
80e755fe | 504 | snapc = ceph_get_snap_context(ci->i_head_snapc); |
1d3576fd SW |
505 | dout(" head snapc %p has %d dirty pages\n", |
506 | snapc, ci->i_wrbuffer_ref_head); | |
1f934b00 YZ |
507 | if (ctl) { |
508 | ctl->i_size = i_size_read(inode); | |
509 | ctl->truncate_size = ci->i_truncate_size; | |
510 | ctl->truncate_seq = ci->i_truncate_seq; | |
511 | ctl->size_stable = false; | |
2a2d927e | 512 | ctl->head_snapc = true; |
1f934b00 | 513 | } |
1d3576fd | 514 | } |
be655596 | 515 | spin_unlock(&ci->i_ceph_lock); |
1d3576fd SW |
516 | return snapc; |
517 | } | |
518 | ||
1f934b00 YZ |
519 | static u64 get_writepages_data_length(struct inode *inode, |
520 | struct page *page, u64 start) | |
521 | { | |
522 | struct ceph_inode_info *ci = ceph_inode(inode); | |
523 | struct ceph_snap_context *snapc = page_snap_context(page); | |
524 | struct ceph_cap_snap *capsnap = NULL; | |
525 | u64 end = i_size_read(inode); | |
526 | ||
527 | if (snapc != ci->i_head_snapc) { | |
528 | bool found = false; | |
529 | spin_lock(&ci->i_ceph_lock); | |
530 | list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { | |
531 | if (capsnap->context == snapc) { | |
532 | if (!capsnap->writing) | |
533 | end = capsnap->size; | |
534 | found = true; | |
535 | break; | |
536 | } | |
537 | } | |
538 | spin_unlock(&ci->i_ceph_lock); | |
539 | WARN_ON(!found); | |
540 | } | |
8ff2d290 JL |
541 | if (end > page_offset(page) + thp_size(page)) |
542 | end = page_offset(page) + thp_size(page); | |
1f934b00 YZ |
543 | return end > start ? end - start : 0; |
544 | } | |
545 | ||
1d3576fd SW |
546 | /* |
547 | * Write a single page, but leave the page locked. | |
548 | * | |
b72b13eb | 549 | * If we get a write error, mark the mapping for error, but still adjust the |
1d3576fd SW |
550 | * dirty page accounting (i.e., page is no longer dirty). |
551 | */ | |
552 | static int writepage_nounlock(struct page *page, struct writeback_control *wbc) | |
553 | { | |
a628304e | 554 | struct folio *folio = page_folio(page); |
6390987f JL |
555 | struct inode *inode = page->mapping->host; |
556 | struct ceph_inode_info *ci = ceph_inode(inode); | |
557 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
6298a337 | 558 | struct ceph_snap_context *snapc, *oldest; |
fc2744aa | 559 | loff_t page_off = page_offset(page); |
6390987f | 560 | int err; |
8ff2d290 | 561 | loff_t len = thp_size(page); |
1f934b00 | 562 | struct ceph_writeback_ctl ceph_wbc; |
6390987f JL |
563 | struct ceph_osd_client *osdc = &fsc->client->osdc; |
564 | struct ceph_osd_request *req; | |
1702e797 | 565 | bool caching = ceph_is_cache_enabled(inode); |
1d3576fd SW |
566 | |
567 | dout("writepage %p idx %lu\n", page, page->index); | |
568 | ||
1d3576fd | 569 | /* verify this is a writeable snap context */ |
61600ef8 | 570 | snapc = page_snap_context(page); |
d37b1d99 | 571 | if (!snapc) { |
1d3576fd | 572 | dout("writepage %p page %p not dirty?\n", inode, page); |
43986881 | 573 | return 0; |
1d3576fd | 574 | } |
05455e11 | 575 | oldest = get_oldest_context(inode, &ceph_wbc, snapc); |
6298a337 | 576 | if (snapc->seq > oldest->seq) { |
1d3576fd | 577 | dout("writepage %p page %p snapc %p not writeable - noop\n", |
61600ef8 | 578 | inode, page, snapc); |
1d3576fd | 579 | /* we should only noop if called by kswapd */ |
fa71fefb | 580 | WARN_ON(!(current->flags & PF_MEMALLOC)); |
6298a337 | 581 | ceph_put_snap_context(oldest); |
fa71fefb | 582 | redirty_page_for_writepage(wbc, page); |
43986881 | 583 | return 0; |
1d3576fd | 584 | } |
6298a337 | 585 | ceph_put_snap_context(oldest); |
1d3576fd SW |
586 | |
587 | /* is this a partial page at end of file? */ | |
1f934b00 | 588 | if (page_off >= ceph_wbc.i_size) { |
a628304e MWO |
589 | dout("folio at %lu beyond eof %llu\n", folio->index, |
590 | ceph_wbc.i_size); | |
591 | folio_invalidate(folio, 0, folio_size(folio)); | |
43986881 | 592 | return 0; |
fc2744aa | 593 | } |
43986881 | 594 | |
1f934b00 YZ |
595 | if (ceph_wbc.i_size < page_off + len) |
596 | len = ceph_wbc.i_size - page_off; | |
1d3576fd | 597 | |
6390987f | 598 | dout("writepage %p page %p index %lu on %llu~%llu snapc %p seq %lld\n", |
1c0a9c2d | 599 | inode, page, page->index, page_off, len, snapc, snapc->seq); |
1d3576fd | 600 | |
314c4737 | 601 | if (atomic_long_inc_return(&fsc->writeback_count) > |
3d14c5d2 | 602 | CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb)) |
503d4fa6 | 603 | fsc->write_congested = true; |
2baba250 | 604 | |
6390987f JL |
605 | req = ceph_osdc_new_request(osdc, &ci->i_layout, ceph_vino(inode), page_off, &len, 0, 1, |
606 | CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, snapc, | |
607 | ceph_wbc.truncate_seq, ceph_wbc.truncate_size, | |
608 | true); | |
3459bd0c XL |
609 | if (IS_ERR(req)) { |
610 | redirty_page_for_writepage(wbc, page); | |
6390987f | 611 | return PTR_ERR(req); |
3459bd0c | 612 | } |
1702e797 JL |
613 | |
614 | set_page_writeback(page); | |
615 | if (caching) | |
616 | ceph_set_page_fscache(page); | |
617 | ceph_fscache_write_to_cache(inode, page_off, len, caching); | |
6390987f JL |
618 | |
619 | /* it may be a short write due to an object boundary */ | |
8ff2d290 | 620 | WARN_ON_ONCE(len > thp_size(page)); |
6390987f JL |
621 | osd_req_op_extent_osd_data_pages(req, 0, &page, len, 0, false, false); |
622 | dout("writepage %llu~%llu (%llu bytes)\n", page_off, len, len); | |
623 | ||
624 | req->r_mtime = inode->i_mtime; | |
625 | err = ceph_osdc_start_request(osdc, req, true); | |
626 | if (!err) | |
627 | err = ceph_osdc_wait_request(osdc, req); | |
628 | ||
8ae99ae2 | 629 | ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency, |
903f4fec | 630 | req->r_end_latency, len, err); |
6390987f JL |
631 | |
632 | ceph_osdc_put_request(req); | |
633 | if (err == 0) | |
634 | err = len; | |
635 | ||
1d3576fd | 636 | if (err < 0) { |
ad15ec06 YZ |
637 | struct writeback_control tmp_wbc; |
638 | if (!wbc) | |
639 | wbc = &tmp_wbc; | |
640 | if (err == -ERESTARTSYS) { | |
641 | /* killed by SIGKILL */ | |
642 | dout("writepage interrupted page %p\n", page); | |
643 | redirty_page_for_writepage(wbc, page); | |
644 | end_page_writeback(page); | |
43986881 | 645 | return err; |
ad15ec06 | 646 | } |
0b98acd6 ID |
647 | if (err == -EBLOCKLISTED) |
648 | fsc->blocklisted = true; | |
ad15ec06 YZ |
649 | dout("writepage setting page/mapping error %d %p\n", |
650 | err, page); | |
1d3576fd | 651 | mapping_set_error(&inode->i_data, err); |
ad15ec06 | 652 | wbc->pages_skipped++; |
1d3576fd SW |
653 | } else { |
654 | dout("writepage cleaned page %p\n", page); | |
655 | err = 0; /* vfs expects us to return 0 */ | |
656 | } | |
379fc7fa JL |
657 | oldest = detach_page_private(page); |
658 | WARN_ON_ONCE(oldest != snapc); | |
1d3576fd SW |
659 | end_page_writeback(page); |
660 | ceph_put_wrbuffer_cap_refs(ci, 1, snapc); | |
6298a337 | 661 | ceph_put_snap_context(snapc); /* page's reference */ |
314c4737 YZ |
662 | |
663 | if (atomic_long_dec_return(&fsc->writeback_count) < | |
664 | CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb)) | |
503d4fa6 | 665 | fsc->write_congested = false; |
314c4737 | 666 | |
1d3576fd SW |
667 | return err; |
668 | } | |
669 | ||
670 | static int ceph_writepage(struct page *page, struct writeback_control *wbc) | |
671 | { | |
dbd646a8 YS |
672 | int err; |
673 | struct inode *inode = page->mapping->host; | |
674 | BUG_ON(!inode); | |
70b666c3 | 675 | ihold(inode); |
1702e797 | 676 | |
503d4fa6 N |
677 | if (wbc->sync_mode == WB_SYNC_NONE && |
678 | ceph_inode_to_client(inode)->write_congested) | |
679 | return AOP_WRITEPAGE_ACTIVATE; | |
680 | ||
1702e797 JL |
681 | wait_on_page_fscache(page); |
682 | ||
dbd646a8 | 683 | err = writepage_nounlock(page, wbc); |
ad15ec06 YZ |
684 | if (err == -ERESTARTSYS) { |
685 | /* direct memory reclaimer was killed by SIGKILL. return 0 | |
686 | * to prevent caller from setting mapping/page error */ | |
687 | err = 0; | |
688 | } | |
1d3576fd | 689 | unlock_page(page); |
dbd646a8 | 690 | iput(inode); |
1d3576fd SW |
691 | return err; |
692 | } | |
693 | ||
1d3576fd SW |
694 | /* |
695 | * async writeback completion handler. | |
696 | * | |
697 | * If we get an error, set the mapping error bit, but not the individual | |
698 | * page error bits. | |
699 | */ | |
85e084fe | 700 | static void writepages_finish(struct ceph_osd_request *req) |
1d3576fd SW |
701 | { |
702 | struct inode *inode = req->r_inode; | |
1d3576fd | 703 | struct ceph_inode_info *ci = ceph_inode(inode); |
87060c10 | 704 | struct ceph_osd_data *osd_data; |
1d3576fd | 705 | struct page *page; |
5b64640c YZ |
706 | int num_pages, total_pages = 0; |
707 | int i, j; | |
708 | int rc = req->r_result; | |
1d3576fd SW |
709 | struct ceph_snap_context *snapc = req->r_snapc; |
710 | struct address_space *mapping = inode->i_mapping; | |
3d14c5d2 | 711 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
903f4fec | 712 | unsigned int len = 0; |
5b64640c | 713 | bool remove_page; |
1d3576fd | 714 | |
5b64640c | 715 | dout("writepages_finish %p rc %d\n", inode, rc); |
26544c62 | 716 | if (rc < 0) { |
1d3576fd | 717 | mapping_set_error(mapping, rc); |
26544c62 | 718 | ceph_set_error_write(ci); |
0b98acd6 ID |
719 | if (rc == -EBLOCKLISTED) |
720 | fsc->blocklisted = true; | |
26544c62 JL |
721 | } else { |
722 | ceph_clear_error_write(ci); | |
723 | } | |
5b64640c YZ |
724 | |
725 | /* | |
726 | * We lost the cache cap, need to truncate the page before | |
727 | * it is unlocked, otherwise we'd truncate it later in the | |
728 | * page truncation thread, possibly losing some data that | |
729 | * raced its way in | |
730 | */ | |
731 | remove_page = !(ceph_caps_issued(ci) & | |
732 | (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)); | |
1d3576fd SW |
733 | |
734 | /* clean all pages */ | |
5b64640c | 735 | for (i = 0; i < req->r_num_ops; i++) { |
642d51fb XL |
736 | if (req->r_ops[i].op != CEPH_OSD_OP_WRITE) { |
737 | pr_warn("%s incorrect op %d req %p index %d tid %llu\n", | |
738 | __func__, req->r_ops[i].op, req, i, req->r_tid); | |
5b64640c | 739 | break; |
642d51fb | 740 | } |
e63dc5c7 | 741 | |
5b64640c YZ |
742 | osd_data = osd_req_op_extent_osd_data(req, i); |
743 | BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); | |
903f4fec | 744 | len += osd_data->length; |
5b64640c YZ |
745 | num_pages = calc_pages_for((u64)osd_data->alignment, |
746 | (u64)osd_data->length); | |
747 | total_pages += num_pages; | |
748 | for (j = 0; j < num_pages; j++) { | |
749 | page = osd_data->pages[j]; | |
750 | BUG_ON(!page); | |
751 | WARN_ON(!PageUptodate(page)); | |
752 | ||
753 | if (atomic_long_dec_return(&fsc->writeback_count) < | |
754 | CONGESTION_OFF_THRESH( | |
755 | fsc->mount_options->congestion_kb)) | |
503d4fa6 | 756 | fsc->write_congested = false; |
5b64640c | 757 | |
379fc7fa | 758 | ceph_put_snap_context(detach_page_private(page)); |
5b64640c | 759 | end_page_writeback(page); |
379fc7fa | 760 | dout("unlocking %p\n", page); |
5b64640c YZ |
761 | |
762 | if (remove_page) | |
763 | generic_error_remove_page(inode->i_mapping, | |
764 | page); | |
765 | ||
766 | unlock_page(page); | |
767 | } | |
768 | dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n", | |
769 | inode, osd_data->length, rc >= 0 ? num_pages : 0); | |
e63dc5c7 | 770 | |
96ac9158 | 771 | release_pages(osd_data->pages, num_pages); |
1d3576fd | 772 | } |
1d3576fd | 773 | |
903f4fec XL |
774 | ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency, |
775 | req->r_end_latency, len, rc); | |
776 | ||
5b64640c YZ |
777 | ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc); |
778 | ||
779 | osd_data = osd_req_op_extent_osd_data(req, 0); | |
87060c10 | 780 | if (osd_data->pages_from_pool) |
a0102bda | 781 | mempool_free(osd_data->pages, ceph_wb_pagevec_pool); |
1d3576fd | 782 | else |
87060c10 | 783 | kfree(osd_data->pages); |
1d3576fd SW |
784 | ceph_osdc_put_request(req); |
785 | } | |
786 | ||
1d3576fd SW |
787 | /* |
788 | * initiate async writeback | |
789 | */ | |
790 | static int ceph_writepages_start(struct address_space *mapping, | |
791 | struct writeback_control *wbc) | |
792 | { | |
793 | struct inode *inode = mapping->host; | |
1d3576fd | 794 | struct ceph_inode_info *ci = ceph_inode(inode); |
fc2744aa YZ |
795 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
796 | struct ceph_vino vino = ceph_vino(inode); | |
2a2d927e | 797 | pgoff_t index, start_index, end = -1; |
80e755fe | 798 | struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc; |
1d3576fd | 799 | struct pagevec pvec; |
1d3576fd | 800 | int rc = 0; |
93407472 | 801 | unsigned int wsize = i_blocksize(inode); |
1d3576fd | 802 | struct ceph_osd_request *req = NULL; |
1f934b00 | 803 | struct ceph_writeback_ctl ceph_wbc; |
590e9d98 | 804 | bool should_loop, range_whole = false; |
af9cc401 | 805 | bool done = false; |
1702e797 | 806 | bool caching = ceph_is_cache_enabled(inode); |
1d3576fd | 807 | |
503d4fa6 N |
808 | if (wbc->sync_mode == WB_SYNC_NONE && |
809 | fsc->write_congested) | |
810 | return 0; | |
811 | ||
3fb99d48 | 812 | dout("writepages_start %p (mode=%s)\n", inode, |
1d3576fd SW |
813 | wbc->sync_mode == WB_SYNC_NONE ? "NONE" : |
814 | (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD")); | |
815 | ||
5d6451b1 | 816 | if (ceph_inode_is_shutdown(inode)) { |
6c93df5d YZ |
817 | if (ci->i_wrbuffer_ref > 0) { |
818 | pr_warn_ratelimited( | |
819 | "writepage_start %p %lld forced umount\n", | |
820 | inode, ceph_ino(inode)); | |
821 | } | |
a341d4df | 822 | mapping_set_error(mapping, -EIO); |
1d3576fd SW |
823 | return -EIO; /* we're in a forced umount, don't write! */ |
824 | } | |
95cca2b4 | 825 | if (fsc->mount_options->wsize < wsize) |
3d14c5d2 | 826 | wsize = fsc->mount_options->wsize; |
1d3576fd | 827 | |
86679820 | 828 | pagevec_init(&pvec); |
1d3576fd | 829 | |
590e9d98 | 830 | start_index = wbc->range_cyclic ? mapping->writeback_index : 0; |
2a2d927e | 831 | index = start_index; |
1d3576fd SW |
832 | |
833 | retry: | |
834 | /* find oldest snap context with dirty data */ | |
05455e11 | 835 | snapc = get_oldest_context(inode, &ceph_wbc, NULL); |
1d3576fd SW |
836 | if (!snapc) { |
837 | /* hmm, why does writepages get called when there | |
838 | is no dirty data? */ | |
839 | dout(" no snap context with dirty data?\n"); | |
840 | goto out; | |
841 | } | |
842 | dout(" oldest snapc is %p seq %lld (%d snaps)\n", | |
843 | snapc, snapc->seq, snapc->num_snaps); | |
fc2744aa | 844 | |
2a2d927e YZ |
845 | should_loop = false; |
846 | if (ceph_wbc.head_snapc && snapc != last_snapc) { | |
847 | /* where to start/end? */ | |
848 | if (wbc->range_cyclic) { | |
849 | index = start_index; | |
850 | end = -1; | |
851 | if (index > 0) | |
852 | should_loop = true; | |
853 | dout(" cyclic, start at %lu\n", index); | |
854 | } else { | |
855 | index = wbc->range_start >> PAGE_SHIFT; | |
856 | end = wbc->range_end >> PAGE_SHIFT; | |
857 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) | |
858 | range_whole = true; | |
859 | dout(" not cyclic, %lu to %lu\n", index, end); | |
860 | } | |
861 | } else if (!ceph_wbc.head_snapc) { | |
862 | /* Do not respect wbc->range_{start,end}. Dirty pages | |
863 | * in that range can be associated with newer snapc. | |
864 | * They are not writeable until we write all dirty pages | |
865 | * associated with 'snapc' get written */ | |
1582af2e | 866 | if (index > 0) |
2a2d927e YZ |
867 | should_loop = true; |
868 | dout(" non-head snapc, range whole\n"); | |
1d3576fd | 869 | } |
2a2d927e YZ |
870 | |
871 | ceph_put_snap_context(last_snapc); | |
1d3576fd SW |
872 | last_snapc = snapc; |
873 | ||
af9cc401 | 874 | while (!done && index <= end) { |
5b64640c | 875 | int num_ops = 0, op_idx; |
0e5ecac7 | 876 | unsigned i, pvec_pages, max_pages, locked_pages = 0; |
5b64640c | 877 | struct page **pages = NULL, **data_pages; |
1d3576fd | 878 | struct page *page; |
0e5ecac7 | 879 | pgoff_t strip_unit_end = 0; |
5b64640c | 880 | u64 offset = 0, len = 0; |
a0102bda | 881 | bool from_pool = false; |
1d3576fd | 882 | |
0e5ecac7 | 883 | max_pages = wsize >> PAGE_SHIFT; |
1d3576fd SW |
884 | |
885 | get_more_pages: | |
2e169296 JL |
886 | pvec_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
887 | end, PAGECACHE_TAG_DIRTY); | |
0ed75fc8 | 888 | dout("pagevec_lookup_range_tag got %d\n", pvec_pages); |
1d3576fd SW |
889 | if (!pvec_pages && !locked_pages) |
890 | break; | |
891 | for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) { | |
892 | page = pvec.pages[i]; | |
893 | dout("? %p idx %lu\n", page, page->index); | |
894 | if (locked_pages == 0) | |
895 | lock_page(page); /* first page */ | |
896 | else if (!trylock_page(page)) | |
897 | break; | |
898 | ||
899 | /* only dirty pages, or our accounting breaks */ | |
900 | if (unlikely(!PageDirty(page)) || | |
901 | unlikely(page->mapping != mapping)) { | |
902 | dout("!dirty or !mapping %p\n", page); | |
903 | unlock_page(page); | |
0713e5f2 | 904 | continue; |
1d3576fd | 905 | } |
af9cc401 YZ |
906 | /* only if matching snap context */ |
907 | pgsnapc = page_snap_context(page); | |
908 | if (pgsnapc != snapc) { | |
909 | dout("page snapc %p %lld != oldest %p %lld\n", | |
910 | pgsnapc, pgsnapc->seq, snapc, snapc->seq); | |
1582af2e YZ |
911 | if (!should_loop && |
912 | !ceph_wbc.head_snapc && | |
913 | wbc->sync_mode != WB_SYNC_NONE) | |
914 | should_loop = true; | |
1d3576fd | 915 | unlock_page(page); |
af9cc401 | 916 | continue; |
1d3576fd | 917 | } |
1f934b00 | 918 | if (page_offset(page) >= ceph_wbc.i_size) { |
a628304e MWO |
919 | struct folio *folio = page_folio(page); |
920 | ||
921 | dout("folio at %lu beyond eof %llu\n", | |
922 | folio->index, ceph_wbc.i_size); | |
c95f1c5f | 923 | if ((ceph_wbc.size_stable || |
a628304e MWO |
924 | folio_pos(folio) >= i_size_read(inode)) && |
925 | folio_clear_dirty_for_io(folio)) | |
926 | folio_invalidate(folio, 0, | |
927 | folio_size(folio)); | |
928 | folio_unlock(folio); | |
af9cc401 YZ |
929 | continue; |
930 | } | |
931 | if (strip_unit_end && (page->index > strip_unit_end)) { | |
932 | dout("end of strip unit %p\n", page); | |
1d3576fd SW |
933 | unlock_page(page); |
934 | break; | |
935 | } | |
1702e797 | 936 | if (PageWriteback(page) || PageFsCache(page)) { |
0713e5f2 YZ |
937 | if (wbc->sync_mode == WB_SYNC_NONE) { |
938 | dout("%p under writeback\n", page); | |
939 | unlock_page(page); | |
940 | continue; | |
941 | } | |
942 | dout("waiting on writeback %p\n", page); | |
943 | wait_on_page_writeback(page); | |
1702e797 | 944 | wait_on_page_fscache(page); |
1d3576fd SW |
945 | } |
946 | ||
1d3576fd SW |
947 | if (!clear_page_dirty_for_io(page)) { |
948 | dout("%p !clear_page_dirty_for_io\n", page); | |
949 | unlock_page(page); | |
0713e5f2 | 950 | continue; |
1d3576fd SW |
951 | } |
952 | ||
e5975c7c AE |
953 | /* |
954 | * We have something to write. If this is | |
955 | * the first locked page this time through, | |
5b64640c YZ |
956 | * calculate max possinle write size and |
957 | * allocate a page array | |
e5975c7c | 958 | */ |
1d3576fd | 959 | if (locked_pages == 0) { |
5b64640c YZ |
960 | u64 objnum; |
961 | u64 objoff; | |
dccbf080 | 962 | u32 xlen; |
5b64640c | 963 | |
1d3576fd | 964 | /* prepare async write request */ |
e5975c7c | 965 | offset = (u64)page_offset(page); |
dccbf080 ID |
966 | ceph_calc_file_object_mapping(&ci->i_layout, |
967 | offset, wsize, | |
968 | &objnum, &objoff, | |
969 | &xlen); | |
970 | len = xlen; | |
8c71897b | 971 | |
3fb99d48 | 972 | num_ops = 1; |
5b64640c | 973 | strip_unit_end = page->index + |
09cbfeaf | 974 | ((len - 1) >> PAGE_SHIFT); |
88486957 | 975 | |
5b64640c | 976 | BUG_ON(pages); |
88486957 | 977 | max_pages = calc_pages_for(0, (u64)len); |
6da2ec56 KC |
978 | pages = kmalloc_array(max_pages, |
979 | sizeof(*pages), | |
980 | GFP_NOFS); | |
88486957 | 981 | if (!pages) { |
a0102bda JL |
982 | from_pool = true; |
983 | pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS); | |
e5975c7c | 984 | BUG_ON(!pages); |
88486957 | 985 | } |
5b64640c YZ |
986 | |
987 | len = 0; | |
988 | } else if (page->index != | |
09cbfeaf | 989 | (offset + len) >> PAGE_SHIFT) { |
a0102bda JL |
990 | if (num_ops >= (from_pool ? CEPH_OSD_SLAB_OPS : |
991 | CEPH_OSD_MAX_OPS)) { | |
5b64640c YZ |
992 | redirty_page_for_writepage(wbc, page); |
993 | unlock_page(page); | |
994 | break; | |
995 | } | |
996 | ||
997 | num_ops++; | |
998 | offset = (u64)page_offset(page); | |
999 | len = 0; | |
1d3576fd SW |
1000 | } |
1001 | ||
1002 | /* note position of first page in pvec */ | |
1d3576fd SW |
1003 | dout("%p will write page %p idx %lu\n", |
1004 | inode, page, page->index); | |
2baba250 | 1005 | |
5b64640c YZ |
1006 | if (atomic_long_inc_return(&fsc->writeback_count) > |
1007 | CONGESTION_ON_THRESH( | |
503d4fa6 N |
1008 | fsc->mount_options->congestion_kb)) |
1009 | fsc->write_congested = true; | |
0713e5f2 YZ |
1010 | |
1011 | pages[locked_pages++] = page; | |
1012 | pvec.pages[i] = NULL; | |
1013 | ||
8ff2d290 | 1014 | len += thp_size(page); |
1d3576fd SW |
1015 | } |
1016 | ||
1017 | /* did we get anything? */ | |
1018 | if (!locked_pages) | |
1019 | goto release_pvec_pages; | |
1020 | if (i) { | |
0713e5f2 YZ |
1021 | unsigned j, n = 0; |
1022 | /* shift unused page to beginning of pvec */ | |
1023 | for (j = 0; j < pvec_pages; j++) { | |
1024 | if (!pvec.pages[j]) | |
1025 | continue; | |
1026 | if (n < j) | |
1027 | pvec.pages[n] = pvec.pages[j]; | |
1028 | n++; | |
1029 | } | |
1030 | pvec.nr = n; | |
1d3576fd SW |
1031 | |
1032 | if (pvec_pages && i == pvec_pages && | |
1033 | locked_pages < max_pages) { | |
1034 | dout("reached end pvec, trying for more\n"); | |
0713e5f2 | 1035 | pagevec_release(&pvec); |
1d3576fd SW |
1036 | goto get_more_pages; |
1037 | } | |
1d3576fd SW |
1038 | } |
1039 | ||
5b64640c | 1040 | new_request: |
e5975c7c | 1041 | offset = page_offset(pages[0]); |
5b64640c YZ |
1042 | len = wsize; |
1043 | ||
1044 | req = ceph_osdc_new_request(&fsc->client->osdc, | |
1045 | &ci->i_layout, vino, | |
1046 | offset, &len, 0, num_ops, | |
1f934b00 YZ |
1047 | CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, |
1048 | snapc, ceph_wbc.truncate_seq, | |
1049 | ceph_wbc.truncate_size, false); | |
5b64640c YZ |
1050 | if (IS_ERR(req)) { |
1051 | req = ceph_osdc_new_request(&fsc->client->osdc, | |
1052 | &ci->i_layout, vino, | |
1053 | offset, &len, 0, | |
1054 | min(num_ops, | |
1055 | CEPH_OSD_SLAB_OPS), | |
1056 | CEPH_OSD_OP_WRITE, | |
54ea0046 | 1057 | CEPH_OSD_FLAG_WRITE, |
1f934b00 YZ |
1058 | snapc, ceph_wbc.truncate_seq, |
1059 | ceph_wbc.truncate_size, true); | |
5b64640c | 1060 | BUG_ON(IS_ERR(req)); |
e1966b49 | 1061 | } |
5b64640c | 1062 | BUG_ON(len < page_offset(pages[locked_pages - 1]) + |
8ff2d290 | 1063 | thp_size(page) - offset); |
5b64640c YZ |
1064 | |
1065 | req->r_callback = writepages_finish; | |
1066 | req->r_inode = inode; | |
1d3576fd | 1067 | |
5b64640c YZ |
1068 | /* Format the osd request message and submit the write */ |
1069 | len = 0; | |
1070 | data_pages = pages; | |
1071 | op_idx = 0; | |
1072 | for (i = 0; i < locked_pages; i++) { | |
1073 | u64 cur_offset = page_offset(pages[i]); | |
1702e797 JL |
1074 | /* |
1075 | * Discontinuity in page range? Ceph can handle that by just passing | |
1076 | * multiple extents in the write op. | |
1077 | */ | |
5b64640c | 1078 | if (offset + len != cur_offset) { |
1702e797 | 1079 | /* If it's full, stop here */ |
3fb99d48 | 1080 | if (op_idx + 1 == req->r_num_ops) |
5b64640c | 1081 | break; |
1702e797 JL |
1082 | |
1083 | /* Kick off an fscache write with what we have so far. */ | |
1084 | ceph_fscache_write_to_cache(inode, offset, len, caching); | |
1085 | ||
1086 | /* Start a new extent */ | |
5b64640c YZ |
1087 | osd_req_op_extent_dup_last(req, op_idx, |
1088 | cur_offset - offset); | |
1089 | dout("writepages got pages at %llu~%llu\n", | |
1090 | offset, len); | |
1091 | osd_req_op_extent_osd_data_pages(req, op_idx, | |
1092 | data_pages, len, 0, | |
a0102bda | 1093 | from_pool, false); |
5b64640c | 1094 | osd_req_op_extent_update(req, op_idx, len); |
e5975c7c | 1095 | |
5b64640c | 1096 | len = 0; |
1702e797 | 1097 | offset = cur_offset; |
5b64640c YZ |
1098 | data_pages = pages + i; |
1099 | op_idx++; | |
1100 | } | |
1101 | ||
1102 | set_page_writeback(pages[i]); | |
1702e797 JL |
1103 | if (caching) |
1104 | ceph_set_page_fscache(pages[i]); | |
8ff2d290 | 1105 | len += thp_size(page); |
5b64640c | 1106 | } |
1702e797 | 1107 | ceph_fscache_write_to_cache(inode, offset, len, caching); |
5b64640c | 1108 | |
1f934b00 YZ |
1109 | if (ceph_wbc.size_stable) { |
1110 | len = min(len, ceph_wbc.i_size - offset); | |
5b64640c YZ |
1111 | } else if (i == locked_pages) { |
1112 | /* writepages_finish() clears writeback pages | |
1113 | * according to the data length, so make sure | |
1114 | * data length covers all locked pages */ | |
8ff2d290 | 1115 | u64 min_len = len + 1 - thp_size(page); |
1f934b00 YZ |
1116 | len = get_writepages_data_length(inode, pages[i - 1], |
1117 | offset); | |
5b64640c YZ |
1118 | len = max(len, min_len); |
1119 | } | |
1120 | dout("writepages got pages at %llu~%llu\n", offset, len); | |
e5975c7c | 1121 | |
5b64640c | 1122 | osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len, |
a0102bda | 1123 | 0, from_pool, false); |
5b64640c | 1124 | osd_req_op_extent_update(req, op_idx, len); |
e5975c7c | 1125 | |
5b64640c YZ |
1126 | BUG_ON(op_idx + 1 != req->r_num_ops); |
1127 | ||
a0102bda | 1128 | from_pool = false; |
5b64640c YZ |
1129 | if (i < locked_pages) { |
1130 | BUG_ON(num_ops <= req->r_num_ops); | |
1131 | num_ops -= req->r_num_ops; | |
5b64640c YZ |
1132 | locked_pages -= i; |
1133 | ||
1134 | /* allocate new pages array for next request */ | |
1135 | data_pages = pages; | |
6da2ec56 KC |
1136 | pages = kmalloc_array(locked_pages, sizeof(*pages), |
1137 | GFP_NOFS); | |
5b64640c | 1138 | if (!pages) { |
a0102bda JL |
1139 | from_pool = true; |
1140 | pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS); | |
5b64640c YZ |
1141 | BUG_ON(!pages); |
1142 | } | |
1143 | memcpy(pages, data_pages + i, | |
1144 | locked_pages * sizeof(*pages)); | |
1145 | memset(data_pages + i, 0, | |
1146 | locked_pages * sizeof(*pages)); | |
1147 | } else { | |
1148 | BUG_ON(num_ops != req->r_num_ops); | |
1149 | index = pages[i - 1]->index + 1; | |
1150 | /* request message now owns the pages array */ | |
1151 | pages = NULL; | |
1152 | } | |
e5975c7c | 1153 | |
fac02ddf | 1154 | req->r_mtime = inode->i_mtime; |
9d6fcb08 SW |
1155 | rc = ceph_osdc_start_request(&fsc->client->osdc, req, true); |
1156 | BUG_ON(rc); | |
1d3576fd SW |
1157 | req = NULL; |
1158 | ||
5b64640c YZ |
1159 | wbc->nr_to_write -= i; |
1160 | if (pages) | |
1161 | goto new_request; | |
1162 | ||
2a2d927e YZ |
1163 | /* |
1164 | * We stop writing back only if we are not doing | |
1165 | * integrity sync. In case of integrity sync we have to | |
1166 | * keep going until we have written all the pages | |
1167 | * we tagged for writeback prior to entering this loop. | |
1168 | */ | |
1169 | if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) | |
af9cc401 | 1170 | done = true; |
1d3576fd SW |
1171 | |
1172 | release_pvec_pages: | |
1173 | dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr, | |
1174 | pvec.nr ? pvec.pages[0] : NULL); | |
1175 | pagevec_release(&pvec); | |
1d3576fd SW |
1176 | } |
1177 | ||
1178 | if (should_loop && !done) { | |
1179 | /* more to do; loop back to beginning of file */ | |
1180 | dout("writepages looping back to beginning of file\n"); | |
2a2d927e | 1181 | end = start_index - 1; /* OK even when start_index == 0 */ |
f275635e YZ |
1182 | |
1183 | /* to write dirty pages associated with next snapc, | |
1184 | * we need to wait until current writes complete */ | |
1185 | if (wbc->sync_mode != WB_SYNC_NONE && | |
1186 | start_index == 0 && /* all dirty pages were checked */ | |
1187 | !ceph_wbc.head_snapc) { | |
1188 | struct page *page; | |
1189 | unsigned i, nr; | |
1190 | index = 0; | |
1191 | while ((index <= end) && | |
1192 | (nr = pagevec_lookup_tag(&pvec, mapping, &index, | |
67fd707f | 1193 | PAGECACHE_TAG_WRITEBACK))) { |
f275635e YZ |
1194 | for (i = 0; i < nr; i++) { |
1195 | page = pvec.pages[i]; | |
1196 | if (page_snap_context(page) != snapc) | |
1197 | continue; | |
1198 | wait_on_page_writeback(page); | |
1199 | } | |
1200 | pagevec_release(&pvec); | |
1201 | cond_resched(); | |
1202 | } | |
1203 | } | |
1204 | ||
2a2d927e | 1205 | start_index = 0; |
1d3576fd SW |
1206 | index = 0; |
1207 | goto retry; | |
1208 | } | |
1209 | ||
1210 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) | |
1211 | mapping->writeback_index = index; | |
1212 | ||
1213 | out: | |
3ed97d63 | 1214 | ceph_osdc_put_request(req); |
2a2d927e YZ |
1215 | ceph_put_snap_context(last_snapc); |
1216 | dout("writepages dend - startone, rc = %d\n", rc); | |
1d3576fd SW |
1217 | return rc; |
1218 | } | |
1219 | ||
1220 | ||
1221 | ||
1222 | /* | |
1223 | * See if a given @snapc is either writeable, or already written. | |
1224 | */ | |
1225 | static int context_is_writeable_or_written(struct inode *inode, | |
1226 | struct ceph_snap_context *snapc) | |
1227 | { | |
05455e11 | 1228 | struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL); |
6298a337 SW |
1229 | int ret = !oldest || snapc->seq <= oldest->seq; |
1230 | ||
1231 | ceph_put_snap_context(oldest); | |
1232 | return ret; | |
1d3576fd SW |
1233 | } |
1234 | ||
18d620f0 JL |
1235 | /** |
1236 | * ceph_find_incompatible - find an incompatible context and return it | |
18d620f0 | 1237 | * @page: page being dirtied |
8f883c24 | 1238 | * |
18d620f0 JL |
1239 | * We are only allowed to write into/dirty a page if the page is |
1240 | * clean, or already dirty within the same snap context. Returns a | |
1241 | * conflicting context if there is one, NULL if there isn't, or a | |
1242 | * negative error code on other errors. | |
1243 | * | |
1244 | * Must be called with page lock held. | |
1d3576fd | 1245 | */ |
18d620f0 | 1246 | static struct ceph_snap_context * |
d45156bf | 1247 | ceph_find_incompatible(struct page *page) |
1d3576fd | 1248 | { |
d45156bf | 1249 | struct inode *inode = page->mapping->host; |
1d3576fd | 1250 | struct ceph_inode_info *ci = ceph_inode(inode); |
1d3576fd | 1251 | |
5d6451b1 JL |
1252 | if (ceph_inode_is_shutdown(inode)) { |
1253 | dout(" page %p %llx:%llx is shutdown\n", page, | |
1254 | ceph_vinop(inode)); | |
1255 | return ERR_PTR(-ESTALE); | |
6c93df5d YZ |
1256 | } |
1257 | ||
18d620f0 JL |
1258 | for (;;) { |
1259 | struct ceph_snap_context *snapc, *oldest; | |
1260 | ||
1261 | wait_on_page_writeback(page); | |
1262 | ||
1263 | snapc = page_snap_context(page); | |
1264 | if (!snapc || snapc == ci->i_head_snapc) | |
1265 | break; | |
1d3576fd | 1266 | |
1d3576fd SW |
1267 | /* |
1268 | * this page is already dirty in another (older) snap | |
1269 | * context! is it writeable now? | |
1270 | */ | |
05455e11 | 1271 | oldest = get_oldest_context(inode, NULL, NULL); |
80e755fe | 1272 | if (snapc->seq > oldest->seq) { |
18d620f0 | 1273 | /* not writeable -- return it for the caller to deal with */ |
6298a337 | 1274 | ceph_put_snap_context(oldest); |
18d620f0 JL |
1275 | dout(" page %p snapc %p not current or oldest\n", page, snapc); |
1276 | return ceph_get_snap_context(snapc); | |
1d3576fd | 1277 | } |
6298a337 | 1278 | ceph_put_snap_context(oldest); |
1d3576fd SW |
1279 | |
1280 | /* yay, writeable, do it now (without dropping page lock) */ | |
18d620f0 JL |
1281 | dout(" page %p snapc %p not current, but oldest\n", page, snapc); |
1282 | if (clear_page_dirty_for_io(page)) { | |
1283 | int r = writepage_nounlock(page, NULL); | |
1284 | if (r < 0) | |
1285 | return ERR_PTR(r); | |
1286 | } | |
1287 | } | |
1288 | return NULL; | |
1289 | } | |
1290 | ||
d801327d | 1291 | static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len, |
78525c74 | 1292 | struct folio *folio, void **_fsdata) |
d801327d JL |
1293 | { |
1294 | struct inode *inode = file_inode(file); | |
1295 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1296 | struct ceph_snap_context *snapc; | |
1297 | ||
78525c74 | 1298 | snapc = ceph_find_incompatible(folio_page(folio, 0)); |
d801327d JL |
1299 | if (snapc) { |
1300 | int r; | |
1301 | ||
78525c74 DH |
1302 | folio_unlock(folio); |
1303 | folio_put(folio); | |
d801327d JL |
1304 | if (IS_ERR(snapc)) |
1305 | return PTR_ERR(snapc); | |
1306 | ||
1307 | ceph_queue_writeback(inode); | |
1308 | r = wait_event_killable(ci->i_cap_wq, | |
1309 | context_is_writeable_or_written(inode, snapc)); | |
1310 | ceph_put_snap_context(snapc); | |
1311 | return r == 0 ? -EAGAIN : r; | |
1312 | } | |
1313 | return 0; | |
1314 | } | |
1315 | ||
18d620f0 JL |
1316 | /* |
1317 | * We are only allowed to write into/dirty the page if the page is | |
1318 | * clean, or already dirty within the same snap context. | |
18d620f0 | 1319 | */ |
1cc16990 | 1320 | static int ceph_write_begin(struct file *file, struct address_space *mapping, |
9d6b0cd7 | 1321 | loff_t pos, unsigned len, |
1cc16990 | 1322 | struct page **pagep, void **fsdata) |
18d620f0 JL |
1323 | { |
1324 | struct inode *inode = file_inode(file); | |
78525c74 | 1325 | struct folio *folio = NULL; |
d801327d | 1326 | int r; |
1d3576fd | 1327 | |
de2a9311 | 1328 | r = netfs_write_begin(file, inode->i_mapping, pos, len, &folio, NULL); |
d801327d | 1329 | if (r == 0) |
78525c74 | 1330 | folio_wait_fscache(folio); |
1cc16990 | 1331 | if (r < 0) { |
78525c74 DH |
1332 | if (folio) |
1333 | folio_put(folio); | |
1cc16990 | 1334 | } else { |
78525c74 DH |
1335 | WARN_ON_ONCE(!folio_test_locked(folio)); |
1336 | *pagep = &folio->page; | |
1cc16990 | 1337 | } |
4af6b225 YS |
1338 | return r; |
1339 | } | |
1340 | ||
1d3576fd SW |
1341 | /* |
1342 | * we don't do anything in here that simple_write_end doesn't do | |
5dda377c | 1343 | * except adjust dirty page accounting |
1d3576fd SW |
1344 | */ |
1345 | static int ceph_write_end(struct file *file, struct address_space *mapping, | |
1346 | loff_t pos, unsigned len, unsigned copied, | |
78525c74 | 1347 | struct page *subpage, void *fsdata) |
1d3576fd | 1348 | { |
78525c74 | 1349 | struct folio *folio = page_folio(subpage); |
496ad9aa | 1350 | struct inode *inode = file_inode(file); |
efb0ca76 | 1351 | bool check_cap = false; |
1d3576fd | 1352 | |
78525c74 DH |
1353 | dout("write_end file %p inode %p folio %p %d~%d (%d)\n", file, |
1354 | inode, folio, (int)pos, (int)copied, (int)len); | |
1d3576fd | 1355 | |
78525c74 | 1356 | if (!folio_test_uptodate(folio)) { |
ce3a8732 | 1357 | /* just return that nothing was copied on a short copy */ |
b9de313c AV |
1358 | if (copied < len) { |
1359 | copied = 0; | |
1360 | goto out; | |
1361 | } | |
78525c74 | 1362 | folio_mark_uptodate(folio); |
b9de313c | 1363 | } |
1d3576fd SW |
1364 | |
1365 | /* did file size increase? */ | |
99c88e69 | 1366 | if (pos+copied > i_size_read(inode)) |
1d3576fd SW |
1367 | check_cap = ceph_inode_set_size(inode, pos+copied); |
1368 | ||
78525c74 | 1369 | folio_mark_dirty(folio); |
1d3576fd | 1370 | |
b9de313c | 1371 | out: |
78525c74 DH |
1372 | folio_unlock(folio); |
1373 | folio_put(folio); | |
1d3576fd SW |
1374 | |
1375 | if (check_cap) | |
1376 | ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL); | |
1377 | ||
1378 | return copied; | |
1379 | } | |
1380 | ||
1d3576fd | 1381 | const struct address_space_operations ceph_aops = { |
6c62371b | 1382 | .read_folio = netfs_read_folio, |
bc899ee1 | 1383 | .readahead = netfs_readahead, |
1d3576fd SW |
1384 | .writepage = ceph_writepage, |
1385 | .writepages = ceph_writepages_start, | |
1386 | .write_begin = ceph_write_begin, | |
1387 | .write_end = ceph_write_end, | |
8fb72b4a | 1388 | .dirty_folio = ceph_dirty_folio, |
9872f4de | 1389 | .invalidate_folio = ceph_invalidate_folio, |
5e414655 | 1390 | .release_folio = ceph_release_folio, |
9c43ff44 | 1391 | .direct_IO = noop_direct_IO, |
1d3576fd SW |
1392 | }; |
1393 | ||
4f7e89f6 YZ |
1394 | static void ceph_block_sigs(sigset_t *oldset) |
1395 | { | |
1396 | sigset_t mask; | |
1397 | siginitsetinv(&mask, sigmask(SIGKILL)); | |
1398 | sigprocmask(SIG_BLOCK, &mask, oldset); | |
1399 | } | |
1400 | ||
1401 | static void ceph_restore_sigs(sigset_t *oldset) | |
1402 | { | |
1403 | sigprocmask(SIG_SETMASK, oldset, NULL); | |
1404 | } | |
1d3576fd SW |
1405 | |
1406 | /* | |
1407 | * vm ops | |
1408 | */ | |
24499847 | 1409 | static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf) |
61f68816 | 1410 | { |
11bac800 | 1411 | struct vm_area_struct *vma = vmf->vma; |
61f68816 YZ |
1412 | struct inode *inode = file_inode(vma->vm_file); |
1413 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1414 | struct ceph_file_info *fi = vma->vm_file->private_data; | |
c403c3a2 | 1415 | loff_t off = (loff_t)vmf->pgoff << PAGE_SHIFT; |
24499847 | 1416 | int want, got, err; |
4f7e89f6 | 1417 | sigset_t oldset; |
24499847 | 1418 | vm_fault_t ret = VM_FAULT_SIGBUS; |
4f7e89f6 | 1419 | |
5d6451b1 JL |
1420 | if (ceph_inode_is_shutdown(inode)) |
1421 | return ret; | |
1422 | ||
4f7e89f6 | 1423 | ceph_block_sigs(&oldset); |
61f68816 | 1424 | |
8ff2d290 JL |
1425 | dout("filemap_fault %p %llx.%llx %llu trying to get caps\n", |
1426 | inode, ceph_vinop(inode), off); | |
61f68816 YZ |
1427 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
1428 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
1429 | else | |
1430 | want = CEPH_CAP_FILE_CACHE; | |
4f7e89f6 YZ |
1431 | |
1432 | got = 0; | |
e72968e1 | 1433 | err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_RD, want, -1, &got); |
24499847 | 1434 | if (err < 0) |
4f7e89f6 | 1435 | goto out_restore; |
6ce026e4 | 1436 | |
8ff2d290 JL |
1437 | dout("filemap_fault %p %llu got cap refs on %s\n", |
1438 | inode, off, ceph_cap_string(got)); | |
61f68816 | 1439 | |
83701246 | 1440 | if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) || |
2b1ac852 | 1441 | ci->i_inline_version == CEPH_INLINE_NONE) { |
5d988308 YZ |
1442 | CEPH_DEFINE_RW_CONTEXT(rw_ctx, got); |
1443 | ceph_add_rw_context(fi, &rw_ctx); | |
11bac800 | 1444 | ret = filemap_fault(vmf); |
5d988308 | 1445 | ceph_del_rw_context(fi, &rw_ctx); |
8ff2d290 JL |
1446 | dout("filemap_fault %p %llu drop cap refs %s ret %x\n", |
1447 | inode, off, ceph_cap_string(got), ret); | |
2b1ac852 | 1448 | } else |
24499847 | 1449 | err = -EAGAIN; |
61f68816 | 1450 | |
61f68816 YZ |
1451 | ceph_put_cap_refs(ci, got); |
1452 | ||
24499847 | 1453 | if (err != -EAGAIN) |
4f7e89f6 | 1454 | goto out_restore; |
83701246 YZ |
1455 | |
1456 | /* read inline data */ | |
09cbfeaf | 1457 | if (off >= PAGE_SIZE) { |
83701246 YZ |
1458 | /* does not support inline data > PAGE_SIZE */ |
1459 | ret = VM_FAULT_SIGBUS; | |
1460 | } else { | |
83701246 | 1461 | struct address_space *mapping = inode->i_mapping; |
057ba5b2 JK |
1462 | struct page *page; |
1463 | ||
1464 | filemap_invalidate_lock_shared(mapping); | |
1465 | page = find_or_create_page(mapping, 0, | |
1466 | mapping_gfp_constraint(mapping, ~__GFP_FS)); | |
83701246 YZ |
1467 | if (!page) { |
1468 | ret = VM_FAULT_OOM; | |
4f7e89f6 | 1469 | goto out_inline; |
83701246 | 1470 | } |
24499847 | 1471 | err = __ceph_do_getattr(inode, page, |
83701246 | 1472 | CEPH_STAT_CAP_INLINE_DATA, true); |
24499847 | 1473 | if (err < 0 || off >= i_size_read(inode)) { |
83701246 | 1474 | unlock_page(page); |
09cbfeaf | 1475 | put_page(page); |
c64a2b05 | 1476 | ret = vmf_error(err); |
4f7e89f6 | 1477 | goto out_inline; |
83701246 | 1478 | } |
24499847 SJ |
1479 | if (err < PAGE_SIZE) |
1480 | zero_user_segment(page, err, PAGE_SIZE); | |
83701246 YZ |
1481 | else |
1482 | flush_dcache_page(page); | |
1483 | SetPageUptodate(page); | |
1484 | vmf->page = page; | |
1485 | ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED; | |
4f7e89f6 | 1486 | out_inline: |
057ba5b2 | 1487 | filemap_invalidate_unlock_shared(mapping); |
8ff2d290 JL |
1488 | dout("filemap_fault %p %llu read inline data ret %x\n", |
1489 | inode, off, ret); | |
83701246 | 1490 | } |
4f7e89f6 YZ |
1491 | out_restore: |
1492 | ceph_restore_sigs(&oldset); | |
24499847 SJ |
1493 | if (err < 0) |
1494 | ret = vmf_error(err); | |
6ce026e4 | 1495 | |
61f68816 YZ |
1496 | return ret; |
1497 | } | |
1d3576fd | 1498 | |
24499847 | 1499 | static vm_fault_t ceph_page_mkwrite(struct vm_fault *vmf) |
1d3576fd | 1500 | { |
11bac800 | 1501 | struct vm_area_struct *vma = vmf->vma; |
496ad9aa | 1502 | struct inode *inode = file_inode(vma->vm_file); |
61f68816 YZ |
1503 | struct ceph_inode_info *ci = ceph_inode(inode); |
1504 | struct ceph_file_info *fi = vma->vm_file->private_data; | |
f66fd9f0 | 1505 | struct ceph_cap_flush *prealloc_cf; |
61f68816 | 1506 | struct page *page = vmf->page; |
6285bc23 | 1507 | loff_t off = page_offset(page); |
61f68816 YZ |
1508 | loff_t size = i_size_read(inode); |
1509 | size_t len; | |
24499847 | 1510 | int want, got, err; |
4f7e89f6 | 1511 | sigset_t oldset; |
24499847 | 1512 | vm_fault_t ret = VM_FAULT_SIGBUS; |
3ca9c3bd | 1513 | |
5d6451b1 JL |
1514 | if (ceph_inode_is_shutdown(inode)) |
1515 | return ret; | |
1516 | ||
f66fd9f0 YZ |
1517 | prealloc_cf = ceph_alloc_cap_flush(); |
1518 | if (!prealloc_cf) | |
6ce026e4 | 1519 | return VM_FAULT_OOM; |
f66fd9f0 | 1520 | |
249c1df5 | 1521 | sb_start_pagefault(inode->i_sb); |
4f7e89f6 | 1522 | ceph_block_sigs(&oldset); |
f66fd9f0 | 1523 | |
8ff2d290 JL |
1524 | if (off + thp_size(page) <= size) |
1525 | len = thp_size(page); | |
1d3576fd | 1526 | else |
8ff2d290 | 1527 | len = offset_in_thp(page, size); |
1d3576fd | 1528 | |
61f68816 YZ |
1529 | dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n", |
1530 | inode, ceph_vinop(inode), off, len, size); | |
1531 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1532 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1533 | else | |
1534 | want = CEPH_CAP_FILE_BUFFER; | |
4f7e89f6 YZ |
1535 | |
1536 | got = 0; | |
e72968e1 | 1537 | err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_WR, want, off + len, &got); |
24499847 | 1538 | if (err < 0) |
4f7e89f6 | 1539 | goto out_free; |
6ce026e4 | 1540 | |
61f68816 YZ |
1541 | dout("page_mkwrite %p %llu~%zd got cap refs on %s\n", |
1542 | inode, off, len, ceph_cap_string(got)); | |
1543 | ||
1544 | /* Update time before taking page lock */ | |
1545 | file_update_time(vma->vm_file); | |
5c308356 | 1546 | inode_inc_iversion_raw(inode); |
4af6b225 | 1547 | |
f0b33df5 | 1548 | do { |
d45156bf JL |
1549 | struct ceph_snap_context *snapc; |
1550 | ||
f0b33df5 | 1551 | lock_page(page); |
4af6b225 | 1552 | |
cb03c143 | 1553 | if (page_mkwrite_check_truncate(page, inode) < 0) { |
f0b33df5 YZ |
1554 | unlock_page(page); |
1555 | ret = VM_FAULT_NOPAGE; | |
1556 | break; | |
1557 | } | |
1558 | ||
d45156bf JL |
1559 | snapc = ceph_find_incompatible(page); |
1560 | if (!snapc) { | |
f0b33df5 YZ |
1561 | /* success. we'll keep the page locked. */ |
1562 | set_page_dirty(page); | |
1563 | ret = VM_FAULT_LOCKED; | |
d45156bf JL |
1564 | break; |
1565 | } | |
1566 | ||
1567 | unlock_page(page); | |
1568 | ||
1569 | if (IS_ERR(snapc)) { | |
1570 | ret = VM_FAULT_SIGBUS; | |
1571 | break; | |
f0b33df5 | 1572 | } |
d45156bf JL |
1573 | |
1574 | ceph_queue_writeback(inode); | |
1575 | err = wait_event_killable(ci->i_cap_wq, | |
1576 | context_is_writeable_or_written(inode, snapc)); | |
1577 | ceph_put_snap_context(snapc); | |
1578 | } while (err == 0); | |
4af6b225 | 1579 | |
083db6fd | 1580 | if (ret == VM_FAULT_LOCKED) { |
61f68816 YZ |
1581 | int dirty; |
1582 | spin_lock(&ci->i_ceph_lock); | |
f66fd9f0 YZ |
1583 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
1584 | &prealloc_cf); | |
61f68816 YZ |
1585 | spin_unlock(&ci->i_ceph_lock); |
1586 | if (dirty) | |
1587 | __mark_inode_dirty(inode, dirty); | |
1588 | } | |
1589 | ||
24499847 | 1590 | dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %x\n", |
61f68816 | 1591 | inode, off, len, ceph_cap_string(got), ret); |
a8810cdc | 1592 | ceph_put_cap_refs_async(ci, got); |
f66fd9f0 | 1593 | out_free: |
4f7e89f6 | 1594 | ceph_restore_sigs(&oldset); |
249c1df5 | 1595 | sb_end_pagefault(inode->i_sb); |
f66fd9f0 | 1596 | ceph_free_cap_flush(prealloc_cf); |
24499847 SJ |
1597 | if (err < 0) |
1598 | ret = vmf_error(err); | |
1d3576fd SW |
1599 | return ret; |
1600 | } | |
1601 | ||
31c542a1 YZ |
1602 | void ceph_fill_inline_data(struct inode *inode, struct page *locked_page, |
1603 | char *data, size_t len) | |
1604 | { | |
1605 | struct address_space *mapping = inode->i_mapping; | |
1606 | struct page *page; | |
1607 | ||
1608 | if (locked_page) { | |
1609 | page = locked_page; | |
1610 | } else { | |
1611 | if (i_size_read(inode) == 0) | |
1612 | return; | |
1613 | page = find_or_create_page(mapping, 0, | |
c62d2555 MH |
1614 | mapping_gfp_constraint(mapping, |
1615 | ~__GFP_FS)); | |
31c542a1 YZ |
1616 | if (!page) |
1617 | return; | |
1618 | if (PageUptodate(page)) { | |
1619 | unlock_page(page); | |
09cbfeaf | 1620 | put_page(page); |
31c542a1 YZ |
1621 | return; |
1622 | } | |
1623 | } | |
1624 | ||
0668ff52 | 1625 | dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n", |
31c542a1 YZ |
1626 | inode, ceph_vinop(inode), len, locked_page); |
1627 | ||
1628 | if (len > 0) { | |
1629 | void *kaddr = kmap_atomic(page); | |
1630 | memcpy(kaddr, data, len); | |
1631 | kunmap_atomic(kaddr); | |
1632 | } | |
1633 | ||
1634 | if (page != locked_page) { | |
09cbfeaf KS |
1635 | if (len < PAGE_SIZE) |
1636 | zero_user_segment(page, len, PAGE_SIZE); | |
31c542a1 YZ |
1637 | else |
1638 | flush_dcache_page(page); | |
1639 | ||
1640 | SetPageUptodate(page); | |
1641 | unlock_page(page); | |
09cbfeaf | 1642 | put_page(page); |
31c542a1 YZ |
1643 | } |
1644 | } | |
1645 | ||
083db6fd | 1646 | int ceph_uninline_data(struct file *file) |
28127bdd | 1647 | { |
083db6fd | 1648 | struct inode *inode = file_inode(file); |
28127bdd YZ |
1649 | struct ceph_inode_info *ci = ceph_inode(inode); |
1650 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
825978fd | 1651 | struct ceph_osd_request *req = NULL; |
083db6fd DH |
1652 | struct ceph_cap_flush *prealloc_cf; |
1653 | struct folio *folio = NULL; | |
c38af982 | 1654 | u64 inline_version = CEPH_INLINE_NONE; |
083db6fd | 1655 | struct page *pages[1]; |
28127bdd | 1656 | int err = 0; |
c38af982 | 1657 | u64 len; |
083db6fd | 1658 | |
825978fd XL |
1659 | spin_lock(&ci->i_ceph_lock); |
1660 | inline_version = ci->i_inline_version; | |
1661 | spin_unlock(&ci->i_ceph_lock); | |
1662 | ||
1663 | dout("uninline_data %p %llx.%llx inline_version %llu\n", | |
1664 | inode, ceph_vinop(inode), inline_version); | |
1665 | ||
1666 | if (inline_version == CEPH_INLINE_NONE) | |
1667 | return 0; | |
1668 | ||
083db6fd DH |
1669 | prealloc_cf = ceph_alloc_cap_flush(); |
1670 | if (!prealloc_cf) | |
1671 | return -ENOMEM; | |
1672 | ||
825978fd XL |
1673 | if (inline_version == 1) /* initial version, no data */ |
1674 | goto out_uninline; | |
1675 | ||
083db6fd DH |
1676 | folio = read_mapping_folio(inode->i_mapping, 0, file); |
1677 | if (IS_ERR(folio)) { | |
1678 | err = PTR_ERR(folio); | |
1679 | goto out; | |
1680 | } | |
1681 | ||
1682 | folio_lock(folio); | |
28127bdd | 1683 | |
083db6fd DH |
1684 | len = i_size_read(inode); |
1685 | if (len > folio_size(folio)) | |
1686 | len = folio_size(folio); | |
28127bdd YZ |
1687 | |
1688 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1689 | ceph_vino(inode), 0, &len, 0, 1, | |
54ea0046 | 1690 | CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE, |
34b759b4 | 1691 | NULL, 0, 0, false); |
28127bdd YZ |
1692 | if (IS_ERR(req)) { |
1693 | err = PTR_ERR(req); | |
083db6fd | 1694 | goto out_unlock; |
28127bdd YZ |
1695 | } |
1696 | ||
fac02ddf | 1697 | req->r_mtime = inode->i_mtime; |
28127bdd YZ |
1698 | err = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
1699 | if (!err) | |
1700 | err = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1701 | ceph_osdc_put_request(req); | |
1702 | if (err < 0) | |
083db6fd | 1703 | goto out_unlock; |
28127bdd YZ |
1704 | |
1705 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1706 | ceph_vino(inode), 0, &len, 1, 3, | |
54ea0046 | 1707 | CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, |
34b759b4 ID |
1708 | NULL, ci->i_truncate_seq, |
1709 | ci->i_truncate_size, false); | |
28127bdd YZ |
1710 | if (IS_ERR(req)) { |
1711 | err = PTR_ERR(req); | |
083db6fd | 1712 | goto out_unlock; |
28127bdd YZ |
1713 | } |
1714 | ||
083db6fd DH |
1715 | pages[0] = folio_page(folio, 0); |
1716 | osd_req_op_extent_osd_data_pages(req, 1, pages, len, 0, false, false); | |
28127bdd | 1717 | |
ec137c10 YZ |
1718 | { |
1719 | __le64 xattr_buf = cpu_to_le64(inline_version); | |
1720 | err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR, | |
1721 | "inline_version", &xattr_buf, | |
1722 | sizeof(xattr_buf), | |
1723 | CEPH_OSD_CMPXATTR_OP_GT, | |
1724 | CEPH_OSD_CMPXATTR_MODE_U64); | |
1725 | if (err) | |
083db6fd | 1726 | goto out_put_req; |
ec137c10 YZ |
1727 | } |
1728 | ||
1729 | { | |
1730 | char xattr_buf[32]; | |
1731 | int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf), | |
1732 | "%llu", inline_version); | |
1733 | err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR, | |
1734 | "inline_version", | |
1735 | xattr_buf, xattr_len, 0, 0); | |
1736 | if (err) | |
083db6fd | 1737 | goto out_put_req; |
ec137c10 | 1738 | } |
28127bdd | 1739 | |
fac02ddf | 1740 | req->r_mtime = inode->i_mtime; |
28127bdd YZ |
1741 | err = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
1742 | if (!err) | |
1743 | err = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
97e27aaa | 1744 | |
8ae99ae2 | 1745 | ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency, |
903f4fec | 1746 | req->r_end_latency, len, err); |
97e27aaa | 1747 | |
825978fd | 1748 | out_uninline: |
083db6fd DH |
1749 | if (!err) { |
1750 | int dirty; | |
1751 | ||
1752 | /* Set to CAP_INLINE_NONE and dirty the caps */ | |
1753 | down_read(&fsc->mdsc->snap_rwsem); | |
1754 | spin_lock(&ci->i_ceph_lock); | |
1755 | ci->i_inline_version = CEPH_INLINE_NONE; | |
1756 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, &prealloc_cf); | |
1757 | spin_unlock(&ci->i_ceph_lock); | |
1758 | up_read(&fsc->mdsc->snap_rwsem); | |
1759 | if (dirty) | |
1760 | __mark_inode_dirty(inode, dirty); | |
1761 | } | |
1762 | out_put_req: | |
28127bdd YZ |
1763 | ceph_osdc_put_request(req); |
1764 | if (err == -ECANCELED) | |
1765 | err = 0; | |
083db6fd | 1766 | out_unlock: |
825978fd XL |
1767 | if (folio) { |
1768 | folio_unlock(folio); | |
1769 | folio_put(folio); | |
1770 | } | |
28127bdd | 1771 | out: |
083db6fd | 1772 | ceph_free_cap_flush(prealloc_cf); |
28127bdd YZ |
1773 | dout("uninline_data %p %llx.%llx inline_version %llu = %d\n", |
1774 | inode, ceph_vinop(inode), inline_version, err); | |
1775 | return err; | |
1776 | } | |
1777 | ||
7cbea8dc | 1778 | static const struct vm_operations_struct ceph_vmops = { |
61f68816 | 1779 | .fault = ceph_filemap_fault, |
1d3576fd SW |
1780 | .page_mkwrite = ceph_page_mkwrite, |
1781 | }; | |
1782 | ||
1783 | int ceph_mmap(struct file *file, struct vm_area_struct *vma) | |
1784 | { | |
1785 | struct address_space *mapping = file->f_mapping; | |
1786 | ||
7e0a1265 | 1787 | if (!mapping->a_ops->read_folio) |
1d3576fd | 1788 | return -ENOEXEC; |
1d3576fd | 1789 | vma->vm_ops = &ceph_vmops; |
1d3576fd SW |
1790 | return 0; |
1791 | } | |
10183a69 YZ |
1792 | |
1793 | enum { | |
1794 | POOL_READ = 1, | |
1795 | POOL_WRITE = 2, | |
1796 | }; | |
1797 | ||
779fe0fb YZ |
1798 | static int __ceph_pool_perm_get(struct ceph_inode_info *ci, |
1799 | s64 pool, struct ceph_string *pool_ns) | |
10183a69 | 1800 | { |
874c8ca1 | 1801 | struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->netfs.inode); |
10183a69 YZ |
1802 | struct ceph_mds_client *mdsc = fsc->mdsc; |
1803 | struct ceph_osd_request *rd_req = NULL, *wr_req = NULL; | |
1804 | struct rb_node **p, *parent; | |
1805 | struct ceph_pool_perm *perm; | |
1806 | struct page **pages; | |
779fe0fb | 1807 | size_t pool_ns_len; |
10183a69 YZ |
1808 | int err = 0, err2 = 0, have = 0; |
1809 | ||
1810 | down_read(&mdsc->pool_perm_rwsem); | |
1811 | p = &mdsc->pool_perm_tree.rb_node; | |
1812 | while (*p) { | |
1813 | perm = rb_entry(*p, struct ceph_pool_perm, node); | |
1814 | if (pool < perm->pool) | |
1815 | p = &(*p)->rb_left; | |
1816 | else if (pool > perm->pool) | |
1817 | p = &(*p)->rb_right; | |
1818 | else { | |
779fe0fb YZ |
1819 | int ret = ceph_compare_string(pool_ns, |
1820 | perm->pool_ns, | |
1821 | perm->pool_ns_len); | |
1822 | if (ret < 0) | |
1823 | p = &(*p)->rb_left; | |
1824 | else if (ret > 0) | |
1825 | p = &(*p)->rb_right; | |
1826 | else { | |
1827 | have = perm->perm; | |
1828 | break; | |
1829 | } | |
10183a69 YZ |
1830 | } |
1831 | } | |
1832 | up_read(&mdsc->pool_perm_rwsem); | |
1833 | if (*p) | |
1834 | goto out; | |
1835 | ||
779fe0fb YZ |
1836 | if (pool_ns) |
1837 | dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n", | |
1838 | pool, (int)pool_ns->len, pool_ns->str); | |
1839 | else | |
1840 | dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool); | |
10183a69 YZ |
1841 | |
1842 | down_write(&mdsc->pool_perm_rwsem); | |
779fe0fb | 1843 | p = &mdsc->pool_perm_tree.rb_node; |
10183a69 YZ |
1844 | parent = NULL; |
1845 | while (*p) { | |
1846 | parent = *p; | |
1847 | perm = rb_entry(parent, struct ceph_pool_perm, node); | |
1848 | if (pool < perm->pool) | |
1849 | p = &(*p)->rb_left; | |
1850 | else if (pool > perm->pool) | |
1851 | p = &(*p)->rb_right; | |
1852 | else { | |
779fe0fb YZ |
1853 | int ret = ceph_compare_string(pool_ns, |
1854 | perm->pool_ns, | |
1855 | perm->pool_ns_len); | |
1856 | if (ret < 0) | |
1857 | p = &(*p)->rb_left; | |
1858 | else if (ret > 0) | |
1859 | p = &(*p)->rb_right; | |
1860 | else { | |
1861 | have = perm->perm; | |
1862 | break; | |
1863 | } | |
10183a69 YZ |
1864 | } |
1865 | } | |
1866 | if (*p) { | |
1867 | up_write(&mdsc->pool_perm_rwsem); | |
1868 | goto out; | |
1869 | } | |
1870 | ||
34b759b4 | 1871 | rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL, |
10183a69 YZ |
1872 | 1, false, GFP_NOFS); |
1873 | if (!rd_req) { | |
1874 | err = -ENOMEM; | |
1875 | goto out_unlock; | |
1876 | } | |
1877 | ||
1878 | rd_req->r_flags = CEPH_OSD_FLAG_READ; | |
1879 | osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0); | |
1880 | rd_req->r_base_oloc.pool = pool; | |
779fe0fb YZ |
1881 | if (pool_ns) |
1882 | rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns); | |
d30291b9 | 1883 | ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino); |
10183a69 | 1884 | |
13d1ad16 ID |
1885 | err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS); |
1886 | if (err) | |
1887 | goto out_unlock; | |
10183a69 | 1888 | |
34b759b4 | 1889 | wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL, |
10183a69 YZ |
1890 | 1, false, GFP_NOFS); |
1891 | if (!wr_req) { | |
1892 | err = -ENOMEM; | |
1893 | goto out_unlock; | |
1894 | } | |
1895 | ||
54ea0046 | 1896 | wr_req->r_flags = CEPH_OSD_FLAG_WRITE; |
10183a69 | 1897 | osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL); |
63244fa1 | 1898 | ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc); |
d30291b9 | 1899 | ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid); |
10183a69 | 1900 | |
13d1ad16 ID |
1901 | err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS); |
1902 | if (err) | |
1903 | goto out_unlock; | |
10183a69 YZ |
1904 | |
1905 | /* one page should be large enough for STAT data */ | |
1906 | pages = ceph_alloc_page_vector(1, GFP_KERNEL); | |
1907 | if (IS_ERR(pages)) { | |
1908 | err = PTR_ERR(pages); | |
1909 | goto out_unlock; | |
1910 | } | |
1911 | ||
1912 | osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE, | |
1913 | 0, false, true); | |
10183a69 YZ |
1914 | err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false); |
1915 | ||
874c8ca1 | 1916 | wr_req->r_mtime = ci->netfs.inode.i_mtime; |
10183a69 YZ |
1917 | err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false); |
1918 | ||
1919 | if (!err) | |
1920 | err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req); | |
1921 | if (!err2) | |
1922 | err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req); | |
1923 | ||
1924 | if (err >= 0 || err == -ENOENT) | |
1925 | have |= POOL_READ; | |
131d7eb4 | 1926 | else if (err != -EPERM) { |
0b98acd6 ID |
1927 | if (err == -EBLOCKLISTED) |
1928 | fsc->blocklisted = true; | |
10183a69 | 1929 | goto out_unlock; |
131d7eb4 | 1930 | } |
10183a69 YZ |
1931 | |
1932 | if (err2 == 0 || err2 == -EEXIST) | |
1933 | have |= POOL_WRITE; | |
1934 | else if (err2 != -EPERM) { | |
0b98acd6 ID |
1935 | if (err2 == -EBLOCKLISTED) |
1936 | fsc->blocklisted = true; | |
10183a69 YZ |
1937 | err = err2; |
1938 | goto out_unlock; | |
1939 | } | |
1940 | ||
779fe0fb YZ |
1941 | pool_ns_len = pool_ns ? pool_ns->len : 0; |
1942 | perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS); | |
10183a69 YZ |
1943 | if (!perm) { |
1944 | err = -ENOMEM; | |
1945 | goto out_unlock; | |
1946 | } | |
1947 | ||
1948 | perm->pool = pool; | |
1949 | perm->perm = have; | |
779fe0fb YZ |
1950 | perm->pool_ns_len = pool_ns_len; |
1951 | if (pool_ns_len > 0) | |
1952 | memcpy(perm->pool_ns, pool_ns->str, pool_ns_len); | |
1953 | perm->pool_ns[pool_ns_len] = 0; | |
1954 | ||
10183a69 YZ |
1955 | rb_link_node(&perm->node, parent, p); |
1956 | rb_insert_color(&perm->node, &mdsc->pool_perm_tree); | |
1957 | err = 0; | |
1958 | out_unlock: | |
1959 | up_write(&mdsc->pool_perm_rwsem); | |
1960 | ||
3ed97d63 ID |
1961 | ceph_osdc_put_request(rd_req); |
1962 | ceph_osdc_put_request(wr_req); | |
10183a69 YZ |
1963 | out: |
1964 | if (!err) | |
1965 | err = have; | |
779fe0fb YZ |
1966 | if (pool_ns) |
1967 | dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n", | |
1968 | pool, (int)pool_ns->len, pool_ns->str, err); | |
1969 | else | |
1970 | dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err); | |
10183a69 YZ |
1971 | return err; |
1972 | } | |
1973 | ||
5e3ded1b | 1974 | int ceph_pool_perm_check(struct inode *inode, int need) |
10183a69 | 1975 | { |
5e3ded1b | 1976 | struct ceph_inode_info *ci = ceph_inode(inode); |
779fe0fb | 1977 | struct ceph_string *pool_ns; |
5e3ded1b | 1978 | s64 pool; |
10183a69 YZ |
1979 | int ret, flags; |
1980 | ||
e9b22501 JL |
1981 | /* Only need to do this for regular files */ |
1982 | if (!S_ISREG(inode->i_mode)) | |
1983 | return 0; | |
1984 | ||
80e80fbb YZ |
1985 | if (ci->i_vino.snap != CEPH_NOSNAP) { |
1986 | /* | |
1987 | * Pool permission check needs to write to the first object. | |
1988 | * But for snapshot, head of the first object may have alread | |
1989 | * been deleted. Skip check to avoid creating orphan object. | |
1990 | */ | |
1991 | return 0; | |
1992 | } | |
1993 | ||
5e3ded1b | 1994 | if (ceph_test_mount_opt(ceph_inode_to_client(inode), |
10183a69 YZ |
1995 | NOPOOLPERM)) |
1996 | return 0; | |
1997 | ||
1998 | spin_lock(&ci->i_ceph_lock); | |
1999 | flags = ci->i_ceph_flags; | |
7627151e | 2000 | pool = ci->i_layout.pool_id; |
10183a69 YZ |
2001 | spin_unlock(&ci->i_ceph_lock); |
2002 | check: | |
2003 | if (flags & CEPH_I_POOL_PERM) { | |
2004 | if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) { | |
7627151e | 2005 | dout("ceph_pool_perm_check pool %lld no read perm\n", |
10183a69 YZ |
2006 | pool); |
2007 | return -EPERM; | |
2008 | } | |
2009 | if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) { | |
7627151e | 2010 | dout("ceph_pool_perm_check pool %lld no write perm\n", |
10183a69 YZ |
2011 | pool); |
2012 | return -EPERM; | |
2013 | } | |
2014 | return 0; | |
2015 | } | |
2016 | ||
779fe0fb YZ |
2017 | pool_ns = ceph_try_get_string(ci->i_layout.pool_ns); |
2018 | ret = __ceph_pool_perm_get(ci, pool, pool_ns); | |
2019 | ceph_put_string(pool_ns); | |
10183a69 YZ |
2020 | if (ret < 0) |
2021 | return ret; | |
2022 | ||
2023 | flags = CEPH_I_POOL_PERM; | |
2024 | if (ret & POOL_READ) | |
2025 | flags |= CEPH_I_POOL_RD; | |
2026 | if (ret & POOL_WRITE) | |
2027 | flags |= CEPH_I_POOL_WR; | |
2028 | ||
2029 | spin_lock(&ci->i_ceph_lock); | |
779fe0fb YZ |
2030 | if (pool == ci->i_layout.pool_id && |
2031 | pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) { | |
2032 | ci->i_ceph_flags |= flags; | |
10183a69 | 2033 | } else { |
7627151e | 2034 | pool = ci->i_layout.pool_id; |
10183a69 YZ |
2035 | flags = ci->i_ceph_flags; |
2036 | } | |
2037 | spin_unlock(&ci->i_ceph_lock); | |
2038 | goto check; | |
2039 | } | |
2040 | ||
2041 | void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc) | |
2042 | { | |
2043 | struct ceph_pool_perm *perm; | |
2044 | struct rb_node *n; | |
2045 | ||
2046 | while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) { | |
2047 | n = rb_first(&mdsc->pool_perm_tree); | |
2048 | perm = rb_entry(n, struct ceph_pool_perm, node); | |
2049 | rb_erase(n, &mdsc->pool_perm_tree); | |
2050 | kfree(perm); | |
2051 | } | |
2052 | } |