]>
Commit | Line | Data |
---|---|---|
1 | // SPDX-License-Identifier: GPL-2.0 | |
2 | #include <linux/ceph/ceph_debug.h> | |
3 | ||
4 | #include <linux/module.h> | |
5 | #include <linux/sched.h> | |
6 | #include <linux/slab.h> | |
7 | #include <linux/file.h> | |
8 | #include <linux/mount.h> | |
9 | #include <linux/namei.h> | |
10 | #include <linux/writeback.h> | |
11 | #include <linux/falloc.h> | |
12 | ||
13 | #include "super.h" | |
14 | #include "mds_client.h" | |
15 | #include "cache.h" | |
16 | ||
17 | static __le32 ceph_flags_sys2wire(u32 flags) | |
18 | { | |
19 | u32 wire_flags = 0; | |
20 | ||
21 | switch (flags & O_ACCMODE) { | |
22 | case O_RDONLY: | |
23 | wire_flags |= CEPH_O_RDONLY; | |
24 | break; | |
25 | case O_WRONLY: | |
26 | wire_flags |= CEPH_O_WRONLY; | |
27 | break; | |
28 | case O_RDWR: | |
29 | wire_flags |= CEPH_O_RDWR; | |
30 | break; | |
31 | } | |
32 | ||
33 | #define ceph_sys2wire(a) if (flags & a) { wire_flags |= CEPH_##a; flags &= ~a; } | |
34 | ||
35 | ceph_sys2wire(O_CREAT); | |
36 | ceph_sys2wire(O_EXCL); | |
37 | ceph_sys2wire(O_TRUNC); | |
38 | ceph_sys2wire(O_DIRECTORY); | |
39 | ceph_sys2wire(O_NOFOLLOW); | |
40 | ||
41 | #undef ceph_sys2wire | |
42 | ||
43 | if (flags) | |
44 | dout("unused open flags: %x", flags); | |
45 | ||
46 | return cpu_to_le32(wire_flags); | |
47 | } | |
48 | ||
49 | /* | |
50 | * Ceph file operations | |
51 | * | |
52 | * Implement basic open/close functionality, and implement | |
53 | * read/write. | |
54 | * | |
55 | * We implement three modes of file I/O: | |
56 | * - buffered uses the generic_file_aio_{read,write} helpers | |
57 | * | |
58 | * - synchronous is used when there is multi-client read/write | |
59 | * sharing, avoids the page cache, and synchronously waits for an | |
60 | * ack from the OSD. | |
61 | * | |
62 | * - direct io takes the variant of the sync path that references | |
63 | * user pages directly. | |
64 | * | |
65 | * fsync() flushes and waits on dirty pages, but just queues metadata | |
66 | * for writeback: since the MDS can recover size and mtime there is no | |
67 | * need to wait for MDS acknowledgement. | |
68 | */ | |
69 | ||
70 | /* | |
71 | * Calculate the length sum of direct io vectors that can | |
72 | * be combined into one page vector. | |
73 | */ | |
74 | static size_t dio_get_pagev_size(const struct iov_iter *it) | |
75 | { | |
76 | const struct iovec *iov = it->iov; | |
77 | const struct iovec *iovend = iov + it->nr_segs; | |
78 | size_t size; | |
79 | ||
80 | size = iov->iov_len - it->iov_offset; | |
81 | /* | |
82 | * An iov can be page vectored when both the current tail | |
83 | * and the next base are page aligned. | |
84 | */ | |
85 | while (PAGE_ALIGNED((iov->iov_base + iov->iov_len)) && | |
86 | (++iov < iovend && PAGE_ALIGNED((iov->iov_base)))) { | |
87 | size += iov->iov_len; | |
88 | } | |
89 | dout("dio_get_pagevlen len = %zu\n", size); | |
90 | return size; | |
91 | } | |
92 | ||
93 | /* | |
94 | * Allocate a page vector based on (@it, @nbytes). | |
95 | * The return value is the tuple describing a page vector, | |
96 | * that is (@pages, @page_align, @num_pages). | |
97 | */ | |
98 | static struct page ** | |
99 | dio_get_pages_alloc(const struct iov_iter *it, size_t nbytes, | |
100 | size_t *page_align, int *num_pages) | |
101 | { | |
102 | struct iov_iter tmp_it = *it; | |
103 | size_t align; | |
104 | struct page **pages; | |
105 | int ret = 0, idx, npages; | |
106 | ||
107 | align = (unsigned long)(it->iov->iov_base + it->iov_offset) & | |
108 | (PAGE_SIZE - 1); | |
109 | npages = calc_pages_for(align, nbytes); | |
110 | pages = kvmalloc(sizeof(*pages) * npages, GFP_KERNEL); | |
111 | if (!pages) | |
112 | return ERR_PTR(-ENOMEM); | |
113 | ||
114 | for (idx = 0; idx < npages; ) { | |
115 | size_t start; | |
116 | ret = iov_iter_get_pages(&tmp_it, pages + idx, nbytes, | |
117 | npages - idx, &start); | |
118 | if (ret < 0) | |
119 | goto fail; | |
120 | ||
121 | iov_iter_advance(&tmp_it, ret); | |
122 | nbytes -= ret; | |
123 | idx += (ret + start + PAGE_SIZE - 1) / PAGE_SIZE; | |
124 | } | |
125 | ||
126 | BUG_ON(nbytes != 0); | |
127 | *num_pages = npages; | |
128 | *page_align = align; | |
129 | dout("dio_get_pages_alloc: got %d pages align %zu\n", npages, align); | |
130 | return pages; | |
131 | fail: | |
132 | ceph_put_page_vector(pages, idx, false); | |
133 | return ERR_PTR(ret); | |
134 | } | |
135 | ||
136 | /* | |
137 | * Prepare an open request. Preallocate ceph_cap to avoid an | |
138 | * inopportune ENOMEM later. | |
139 | */ | |
140 | static struct ceph_mds_request * | |
141 | prepare_open_request(struct super_block *sb, int flags, int create_mode) | |
142 | { | |
143 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); | |
144 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
145 | struct ceph_mds_request *req; | |
146 | int want_auth = USE_ANY_MDS; | |
147 | int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN; | |
148 | ||
149 | if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC)) | |
150 | want_auth = USE_AUTH_MDS; | |
151 | ||
152 | req = ceph_mdsc_create_request(mdsc, op, want_auth); | |
153 | if (IS_ERR(req)) | |
154 | goto out; | |
155 | req->r_fmode = ceph_flags_to_mode(flags); | |
156 | req->r_args.open.flags = ceph_flags_sys2wire(flags); | |
157 | req->r_args.open.mode = cpu_to_le32(create_mode); | |
158 | out: | |
159 | return req; | |
160 | } | |
161 | ||
162 | /* | |
163 | * initialize private struct file data. | |
164 | * if we fail, clean up by dropping fmode reference on the ceph_inode | |
165 | */ | |
166 | static int ceph_init_file(struct inode *inode, struct file *file, int fmode) | |
167 | { | |
168 | struct ceph_file_info *cf; | |
169 | int ret = 0; | |
170 | ||
171 | switch (inode->i_mode & S_IFMT) { | |
172 | case S_IFREG: | |
173 | ceph_fscache_register_inode_cookie(inode); | |
174 | ceph_fscache_file_set_cookie(inode, file); | |
175 | case S_IFDIR: | |
176 | dout("init_file %p %p 0%o (regular)\n", inode, file, | |
177 | inode->i_mode); | |
178 | cf = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL); | |
179 | if (!cf) { | |
180 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
181 | return -ENOMEM; | |
182 | } | |
183 | cf->fmode = fmode; | |
184 | ||
185 | spin_lock_init(&cf->rw_contexts_lock); | |
186 | INIT_LIST_HEAD(&cf->rw_contexts); | |
187 | ||
188 | cf->next_offset = 2; | |
189 | cf->readdir_cache_idx = -1; | |
190 | file->private_data = cf; | |
191 | BUG_ON(inode->i_fop->release != ceph_release); | |
192 | break; | |
193 | ||
194 | case S_IFLNK: | |
195 | dout("init_file %p %p 0%o (symlink)\n", inode, file, | |
196 | inode->i_mode); | |
197 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
198 | break; | |
199 | ||
200 | default: | |
201 | dout("init_file %p %p 0%o (special)\n", inode, file, | |
202 | inode->i_mode); | |
203 | /* | |
204 | * we need to drop the open ref now, since we don't | |
205 | * have .release set to ceph_release. | |
206 | */ | |
207 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
208 | BUG_ON(inode->i_fop->release == ceph_release); | |
209 | ||
210 | /* call the proper open fop */ | |
211 | ret = inode->i_fop->open(inode, file); | |
212 | } | |
213 | return ret; | |
214 | } | |
215 | ||
216 | /* | |
217 | * try renew caps after session gets killed. | |
218 | */ | |
219 | int ceph_renew_caps(struct inode *inode) | |
220 | { | |
221 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; | |
222 | struct ceph_inode_info *ci = ceph_inode(inode); | |
223 | struct ceph_mds_request *req; | |
224 | int err, flags, wanted; | |
225 | ||
226 | spin_lock(&ci->i_ceph_lock); | |
227 | wanted = __ceph_caps_file_wanted(ci); | |
228 | if (__ceph_is_any_real_caps(ci) && | |
229 | (!(wanted & CEPH_CAP_ANY_WR) || ci->i_auth_cap)) { | |
230 | int issued = __ceph_caps_issued(ci, NULL); | |
231 | spin_unlock(&ci->i_ceph_lock); | |
232 | dout("renew caps %p want %s issued %s updating mds_wanted\n", | |
233 | inode, ceph_cap_string(wanted), ceph_cap_string(issued)); | |
234 | ceph_check_caps(ci, 0, NULL); | |
235 | return 0; | |
236 | } | |
237 | spin_unlock(&ci->i_ceph_lock); | |
238 | ||
239 | flags = 0; | |
240 | if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR)) | |
241 | flags = O_RDWR; | |
242 | else if (wanted & CEPH_CAP_FILE_RD) | |
243 | flags = O_RDONLY; | |
244 | else if (wanted & CEPH_CAP_FILE_WR) | |
245 | flags = O_WRONLY; | |
246 | #ifdef O_LAZY | |
247 | if (wanted & CEPH_CAP_FILE_LAZYIO) | |
248 | flags |= O_LAZY; | |
249 | #endif | |
250 | ||
251 | req = prepare_open_request(inode->i_sb, flags, 0); | |
252 | if (IS_ERR(req)) { | |
253 | err = PTR_ERR(req); | |
254 | goto out; | |
255 | } | |
256 | ||
257 | req->r_inode = inode; | |
258 | ihold(inode); | |
259 | req->r_num_caps = 1; | |
260 | req->r_fmode = -1; | |
261 | ||
262 | err = ceph_mdsc_do_request(mdsc, NULL, req); | |
263 | ceph_mdsc_put_request(req); | |
264 | out: | |
265 | dout("renew caps %p open result=%d\n", inode, err); | |
266 | return err < 0 ? err : 0; | |
267 | } | |
268 | ||
269 | /* | |
270 | * If we already have the requisite capabilities, we can satisfy | |
271 | * the open request locally (no need to request new caps from the | |
272 | * MDS). We do, however, need to inform the MDS (asynchronously) | |
273 | * if our wanted caps set expands. | |
274 | */ | |
275 | int ceph_open(struct inode *inode, struct file *file) | |
276 | { | |
277 | struct ceph_inode_info *ci = ceph_inode(inode); | |
278 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); | |
279 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
280 | struct ceph_mds_request *req; | |
281 | struct ceph_file_info *cf = file->private_data; | |
282 | int err; | |
283 | int flags, fmode, wanted; | |
284 | ||
285 | if (cf) { | |
286 | dout("open file %p is already opened\n", file); | |
287 | return 0; | |
288 | } | |
289 | ||
290 | /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */ | |
291 | flags = file->f_flags & ~(O_CREAT|O_EXCL); | |
292 | if (S_ISDIR(inode->i_mode)) | |
293 | flags = O_DIRECTORY; /* mds likes to know */ | |
294 | ||
295 | dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode, | |
296 | ceph_vinop(inode), file, flags, file->f_flags); | |
297 | fmode = ceph_flags_to_mode(flags); | |
298 | wanted = ceph_caps_for_mode(fmode); | |
299 | ||
300 | /* snapped files are read-only */ | |
301 | if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE)) | |
302 | return -EROFS; | |
303 | ||
304 | /* trivially open snapdir */ | |
305 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
306 | spin_lock(&ci->i_ceph_lock); | |
307 | __ceph_get_fmode(ci, fmode); | |
308 | spin_unlock(&ci->i_ceph_lock); | |
309 | return ceph_init_file(inode, file, fmode); | |
310 | } | |
311 | ||
312 | /* | |
313 | * No need to block if we have caps on the auth MDS (for | |
314 | * write) or any MDS (for read). Update wanted set | |
315 | * asynchronously. | |
316 | */ | |
317 | spin_lock(&ci->i_ceph_lock); | |
318 | if (__ceph_is_any_real_caps(ci) && | |
319 | (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) { | |
320 | int mds_wanted = __ceph_caps_mds_wanted(ci, true); | |
321 | int issued = __ceph_caps_issued(ci, NULL); | |
322 | ||
323 | dout("open %p fmode %d want %s issued %s using existing\n", | |
324 | inode, fmode, ceph_cap_string(wanted), | |
325 | ceph_cap_string(issued)); | |
326 | __ceph_get_fmode(ci, fmode); | |
327 | spin_unlock(&ci->i_ceph_lock); | |
328 | ||
329 | /* adjust wanted? */ | |
330 | if ((issued & wanted) != wanted && | |
331 | (mds_wanted & wanted) != wanted && | |
332 | ceph_snap(inode) != CEPH_SNAPDIR) | |
333 | ceph_check_caps(ci, 0, NULL); | |
334 | ||
335 | return ceph_init_file(inode, file, fmode); | |
336 | } else if (ceph_snap(inode) != CEPH_NOSNAP && | |
337 | (ci->i_snap_caps & wanted) == wanted) { | |
338 | __ceph_get_fmode(ci, fmode); | |
339 | spin_unlock(&ci->i_ceph_lock); | |
340 | return ceph_init_file(inode, file, fmode); | |
341 | } | |
342 | ||
343 | spin_unlock(&ci->i_ceph_lock); | |
344 | ||
345 | dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted)); | |
346 | req = prepare_open_request(inode->i_sb, flags, 0); | |
347 | if (IS_ERR(req)) { | |
348 | err = PTR_ERR(req); | |
349 | goto out; | |
350 | } | |
351 | req->r_inode = inode; | |
352 | ihold(inode); | |
353 | ||
354 | req->r_num_caps = 1; | |
355 | err = ceph_mdsc_do_request(mdsc, NULL, req); | |
356 | if (!err) | |
357 | err = ceph_init_file(inode, file, req->r_fmode); | |
358 | ceph_mdsc_put_request(req); | |
359 | dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode)); | |
360 | out: | |
361 | return err; | |
362 | } | |
363 | ||
364 | ||
365 | /* | |
366 | * Do a lookup + open with a single request. If we get a non-existent | |
367 | * file or symlink, return 1 so the VFS can retry. | |
368 | */ | |
369 | int ceph_atomic_open(struct inode *dir, struct dentry *dentry, | |
370 | struct file *file, unsigned flags, umode_t mode, | |
371 | int *opened) | |
372 | { | |
373 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); | |
374 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
375 | struct ceph_mds_request *req; | |
376 | struct dentry *dn; | |
377 | struct ceph_acls_info acls = {}; | |
378 | int mask; | |
379 | int err; | |
380 | ||
381 | dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n", | |
382 | dir, dentry, dentry, | |
383 | d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode); | |
384 | ||
385 | if (dentry->d_name.len > NAME_MAX) | |
386 | return -ENAMETOOLONG; | |
387 | ||
388 | if (flags & O_CREAT) { | |
389 | if (ceph_quota_is_max_files_exceeded(dir)) | |
390 | return -EDQUOT; | |
391 | err = ceph_pre_init_acls(dir, &mode, &acls); | |
392 | if (err < 0) | |
393 | return err; | |
394 | } | |
395 | ||
396 | /* do the open */ | |
397 | req = prepare_open_request(dir->i_sb, flags, mode); | |
398 | if (IS_ERR(req)) { | |
399 | err = PTR_ERR(req); | |
400 | goto out_acl; | |
401 | } | |
402 | req->r_dentry = dget(dentry); | |
403 | req->r_num_caps = 2; | |
404 | if (flags & O_CREAT) { | |
405 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
406 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
407 | if (acls.pagelist) { | |
408 | req->r_pagelist = acls.pagelist; | |
409 | acls.pagelist = NULL; | |
410 | } | |
411 | } | |
412 | ||
413 | mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED; | |
414 | if (ceph_security_xattr_wanted(dir)) | |
415 | mask |= CEPH_CAP_XATTR_SHARED; | |
416 | req->r_args.open.mask = cpu_to_le32(mask); | |
417 | ||
418 | req->r_parent = dir; | |
419 | set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); | |
420 | err = ceph_mdsc_do_request(mdsc, | |
421 | (flags & (O_CREAT|O_TRUNC)) ? dir : NULL, | |
422 | req); | |
423 | err = ceph_handle_snapdir(req, dentry, err); | |
424 | if (err) | |
425 | goto out_req; | |
426 | ||
427 | if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry) | |
428 | err = ceph_handle_notrace_create(dir, dentry); | |
429 | ||
430 | if (d_in_lookup(dentry)) { | |
431 | dn = ceph_finish_lookup(req, dentry, err); | |
432 | if (IS_ERR(dn)) | |
433 | err = PTR_ERR(dn); | |
434 | } else { | |
435 | /* we were given a hashed negative dentry */ | |
436 | dn = NULL; | |
437 | } | |
438 | if (err) | |
439 | goto out_req; | |
440 | if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) { | |
441 | /* make vfs retry on splice, ENOENT, or symlink */ | |
442 | dout("atomic_open finish_no_open on dn %p\n", dn); | |
443 | err = finish_no_open(file, dn); | |
444 | } else { | |
445 | dout("atomic_open finish_open on dn %p\n", dn); | |
446 | if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) { | |
447 | ceph_init_inode_acls(d_inode(dentry), &acls); | |
448 | *opened |= FILE_CREATED; | |
449 | } | |
450 | err = finish_open(file, dentry, ceph_open, opened); | |
451 | } | |
452 | out_req: | |
453 | if (!req->r_err && req->r_target_inode) | |
454 | ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode); | |
455 | ceph_mdsc_put_request(req); | |
456 | out_acl: | |
457 | ceph_release_acls_info(&acls); | |
458 | dout("atomic_open result=%d\n", err); | |
459 | return err; | |
460 | } | |
461 | ||
462 | int ceph_release(struct inode *inode, struct file *file) | |
463 | { | |
464 | struct ceph_inode_info *ci = ceph_inode(inode); | |
465 | struct ceph_file_info *cf = file->private_data; | |
466 | ||
467 | dout("release inode %p file %p\n", inode, file); | |
468 | ceph_put_fmode(ci, cf->fmode); | |
469 | if (cf->last_readdir) | |
470 | ceph_mdsc_put_request(cf->last_readdir); | |
471 | kfree(cf->last_name); | |
472 | kfree(cf->dir_info); | |
473 | WARN_ON(!list_empty(&cf->rw_contexts)); | |
474 | kmem_cache_free(ceph_file_cachep, cf); | |
475 | ||
476 | /* wake up anyone waiting for caps on this inode */ | |
477 | wake_up_all(&ci->i_cap_wq); | |
478 | return 0; | |
479 | } | |
480 | ||
481 | enum { | |
482 | HAVE_RETRIED = 1, | |
483 | CHECK_EOF = 2, | |
484 | READ_INLINE = 3, | |
485 | }; | |
486 | ||
487 | /* | |
488 | * Read a range of bytes striped over one or more objects. Iterate over | |
489 | * objects we stripe over. (That's not atomic, but good enough for now.) | |
490 | * | |
491 | * If we get a short result from the OSD, check against i_size; we need to | |
492 | * only return a short read to the caller if we hit EOF. | |
493 | */ | |
494 | static int striped_read(struct inode *inode, | |
495 | u64 pos, u64 len, | |
496 | struct page **pages, int num_pages, | |
497 | int page_align, int *checkeof) | |
498 | { | |
499 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
500 | struct ceph_inode_info *ci = ceph_inode(inode); | |
501 | u64 this_len; | |
502 | loff_t i_size; | |
503 | int page_idx; | |
504 | int ret, read = 0; | |
505 | bool hit_stripe, was_short; | |
506 | ||
507 | /* | |
508 | * we may need to do multiple reads. not atomic, unfortunately. | |
509 | */ | |
510 | more: | |
511 | this_len = len; | |
512 | page_idx = (page_align + read) >> PAGE_SHIFT; | |
513 | ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), | |
514 | &ci->i_layout, pos, &this_len, | |
515 | ci->i_truncate_seq, ci->i_truncate_size, | |
516 | pages + page_idx, num_pages - page_idx, | |
517 | ((page_align + read) & ~PAGE_MASK)); | |
518 | if (ret == -ENOENT) | |
519 | ret = 0; | |
520 | hit_stripe = this_len < len; | |
521 | was_short = ret >= 0 && ret < this_len; | |
522 | dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, len, read, | |
523 | ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : ""); | |
524 | ||
525 | i_size = i_size_read(inode); | |
526 | if (ret >= 0) { | |
527 | if (was_short && (pos + ret < i_size)) { | |
528 | int zlen = min(this_len - ret, i_size - pos - ret); | |
529 | int zoff = page_align + read + ret; | |
530 | dout(" zero gap %llu to %llu\n", | |
531 | pos + ret, pos + ret + zlen); | |
532 | ceph_zero_page_vector_range(zoff, zlen, pages); | |
533 | ret += zlen; | |
534 | } | |
535 | ||
536 | read += ret; | |
537 | pos += ret; | |
538 | len -= ret; | |
539 | ||
540 | /* hit stripe and need continue*/ | |
541 | if (len && hit_stripe && pos < i_size) | |
542 | goto more; | |
543 | } | |
544 | ||
545 | if (read > 0) { | |
546 | ret = read; | |
547 | /* did we bounce off eof? */ | |
548 | if (pos + len > i_size) | |
549 | *checkeof = CHECK_EOF; | |
550 | } | |
551 | ||
552 | dout("striped_read returns %d\n", ret); | |
553 | return ret; | |
554 | } | |
555 | ||
556 | /* | |
557 | * Completely synchronous read and write methods. Direct from __user | |
558 | * buffer to osd, or directly to user pages (if O_DIRECT). | |
559 | * | |
560 | * If the read spans object boundary, just do multiple reads. | |
561 | */ | |
562 | static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to, | |
563 | int *checkeof) | |
564 | { | |
565 | struct file *file = iocb->ki_filp; | |
566 | struct inode *inode = file_inode(file); | |
567 | struct page **pages; | |
568 | u64 off = iocb->ki_pos; | |
569 | int num_pages; | |
570 | ssize_t ret; | |
571 | size_t len = iov_iter_count(to); | |
572 | ||
573 | dout("sync_read on file %p %llu~%u %s\n", file, off, (unsigned)len, | |
574 | (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); | |
575 | ||
576 | if (!len) | |
577 | return 0; | |
578 | /* | |
579 | * flush any page cache pages in this range. this | |
580 | * will make concurrent normal and sync io slow, | |
581 | * but it will at least behave sensibly when they are | |
582 | * in sequence. | |
583 | */ | |
584 | ret = filemap_write_and_wait_range(inode->i_mapping, off, | |
585 | off + len); | |
586 | if (ret < 0) | |
587 | return ret; | |
588 | ||
589 | if (unlikely(to->type & ITER_PIPE)) { | |
590 | size_t page_off; | |
591 | ret = iov_iter_get_pages_alloc(to, &pages, len, | |
592 | &page_off); | |
593 | if (ret <= 0) | |
594 | return -ENOMEM; | |
595 | num_pages = DIV_ROUND_UP(ret + page_off, PAGE_SIZE); | |
596 | ||
597 | ret = striped_read(inode, off, ret, pages, num_pages, | |
598 | page_off, checkeof); | |
599 | if (ret > 0) { | |
600 | iov_iter_advance(to, ret); | |
601 | off += ret; | |
602 | } else { | |
603 | iov_iter_advance(to, 0); | |
604 | } | |
605 | ceph_put_page_vector(pages, num_pages, false); | |
606 | } else { | |
607 | num_pages = calc_pages_for(off, len); | |
608 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); | |
609 | if (IS_ERR(pages)) | |
610 | return PTR_ERR(pages); | |
611 | ||
612 | ret = striped_read(inode, off, len, pages, num_pages, | |
613 | (off & ~PAGE_MASK), checkeof); | |
614 | if (ret > 0) { | |
615 | int l, k = 0; | |
616 | size_t left = ret; | |
617 | ||
618 | while (left) { | |
619 | size_t page_off = off & ~PAGE_MASK; | |
620 | size_t copy = min_t(size_t, left, | |
621 | PAGE_SIZE - page_off); | |
622 | l = copy_page_to_iter(pages[k++], page_off, | |
623 | copy, to); | |
624 | off += l; | |
625 | left -= l; | |
626 | if (l < copy) | |
627 | break; | |
628 | } | |
629 | } | |
630 | ceph_release_page_vector(pages, num_pages); | |
631 | } | |
632 | ||
633 | if (off > iocb->ki_pos) { | |
634 | ret = off - iocb->ki_pos; | |
635 | iocb->ki_pos = off; | |
636 | } | |
637 | ||
638 | dout("sync_read result %zd\n", ret); | |
639 | return ret; | |
640 | } | |
641 | ||
642 | struct ceph_aio_request { | |
643 | struct kiocb *iocb; | |
644 | size_t total_len; | |
645 | bool write; | |
646 | bool should_dirty; | |
647 | int error; | |
648 | struct list_head osd_reqs; | |
649 | unsigned num_reqs; | |
650 | atomic_t pending_reqs; | |
651 | struct timespec mtime; | |
652 | struct ceph_cap_flush *prealloc_cf; | |
653 | }; | |
654 | ||
655 | struct ceph_aio_work { | |
656 | struct work_struct work; | |
657 | struct ceph_osd_request *req; | |
658 | }; | |
659 | ||
660 | static void ceph_aio_retry_work(struct work_struct *work); | |
661 | ||
662 | static void ceph_aio_complete(struct inode *inode, | |
663 | struct ceph_aio_request *aio_req) | |
664 | { | |
665 | struct ceph_inode_info *ci = ceph_inode(inode); | |
666 | int ret; | |
667 | ||
668 | if (!atomic_dec_and_test(&aio_req->pending_reqs)) | |
669 | return; | |
670 | ||
671 | ret = aio_req->error; | |
672 | if (!ret) | |
673 | ret = aio_req->total_len; | |
674 | ||
675 | dout("ceph_aio_complete %p rc %d\n", inode, ret); | |
676 | ||
677 | if (ret >= 0 && aio_req->write) { | |
678 | int dirty; | |
679 | ||
680 | loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len; | |
681 | if (endoff > i_size_read(inode)) { | |
682 | if (ceph_inode_set_size(inode, endoff)) | |
683 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); | |
684 | } | |
685 | ||
686 | spin_lock(&ci->i_ceph_lock); | |
687 | ci->i_inline_version = CEPH_INLINE_NONE; | |
688 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, | |
689 | &aio_req->prealloc_cf); | |
690 | spin_unlock(&ci->i_ceph_lock); | |
691 | if (dirty) | |
692 | __mark_inode_dirty(inode, dirty); | |
693 | ||
694 | } | |
695 | ||
696 | ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR : | |
697 | CEPH_CAP_FILE_RD)); | |
698 | ||
699 | aio_req->iocb->ki_complete(aio_req->iocb, ret, 0); | |
700 | ||
701 | ceph_free_cap_flush(aio_req->prealloc_cf); | |
702 | kfree(aio_req); | |
703 | } | |
704 | ||
705 | static void ceph_aio_complete_req(struct ceph_osd_request *req) | |
706 | { | |
707 | int rc = req->r_result; | |
708 | struct inode *inode = req->r_inode; | |
709 | struct ceph_aio_request *aio_req = req->r_priv; | |
710 | struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0); | |
711 | int num_pages = calc_pages_for((u64)osd_data->alignment, | |
712 | osd_data->length); | |
713 | ||
714 | dout("ceph_aio_complete_req %p rc %d bytes %llu\n", | |
715 | inode, rc, osd_data->length); | |
716 | ||
717 | if (rc == -EOLDSNAPC) { | |
718 | struct ceph_aio_work *aio_work; | |
719 | BUG_ON(!aio_req->write); | |
720 | ||
721 | aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS); | |
722 | if (aio_work) { | |
723 | INIT_WORK(&aio_work->work, ceph_aio_retry_work); | |
724 | aio_work->req = req; | |
725 | queue_work(ceph_inode_to_client(inode)->inode_wq, | |
726 | &aio_work->work); | |
727 | return; | |
728 | } | |
729 | rc = -ENOMEM; | |
730 | } else if (!aio_req->write) { | |
731 | if (rc == -ENOENT) | |
732 | rc = 0; | |
733 | if (rc >= 0 && osd_data->length > rc) { | |
734 | int zoff = osd_data->alignment + rc; | |
735 | int zlen = osd_data->length - rc; | |
736 | /* | |
737 | * If read is satisfied by single OSD request, | |
738 | * it can pass EOF. Otherwise read is within | |
739 | * i_size. | |
740 | */ | |
741 | if (aio_req->num_reqs == 1) { | |
742 | loff_t i_size = i_size_read(inode); | |
743 | loff_t endoff = aio_req->iocb->ki_pos + rc; | |
744 | if (endoff < i_size) | |
745 | zlen = min_t(size_t, zlen, | |
746 | i_size - endoff); | |
747 | aio_req->total_len = rc + zlen; | |
748 | } | |
749 | ||
750 | if (zlen > 0) | |
751 | ceph_zero_page_vector_range(zoff, zlen, | |
752 | osd_data->pages); | |
753 | } | |
754 | } | |
755 | ||
756 | ceph_put_page_vector(osd_data->pages, num_pages, aio_req->should_dirty); | |
757 | ceph_osdc_put_request(req); | |
758 | ||
759 | if (rc < 0) | |
760 | cmpxchg(&aio_req->error, 0, rc); | |
761 | ||
762 | ceph_aio_complete(inode, aio_req); | |
763 | return; | |
764 | } | |
765 | ||
766 | static void ceph_aio_retry_work(struct work_struct *work) | |
767 | { | |
768 | struct ceph_aio_work *aio_work = | |
769 | container_of(work, struct ceph_aio_work, work); | |
770 | struct ceph_osd_request *orig_req = aio_work->req; | |
771 | struct ceph_aio_request *aio_req = orig_req->r_priv; | |
772 | struct inode *inode = orig_req->r_inode; | |
773 | struct ceph_inode_info *ci = ceph_inode(inode); | |
774 | struct ceph_snap_context *snapc; | |
775 | struct ceph_osd_request *req; | |
776 | int ret; | |
777 | ||
778 | spin_lock(&ci->i_ceph_lock); | |
779 | if (__ceph_have_pending_cap_snap(ci)) { | |
780 | struct ceph_cap_snap *capsnap = | |
781 | list_last_entry(&ci->i_cap_snaps, | |
782 | struct ceph_cap_snap, | |
783 | ci_item); | |
784 | snapc = ceph_get_snap_context(capsnap->context); | |
785 | } else { | |
786 | BUG_ON(!ci->i_head_snapc); | |
787 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
788 | } | |
789 | spin_unlock(&ci->i_ceph_lock); | |
790 | ||
791 | req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 2, | |
792 | false, GFP_NOFS); | |
793 | if (!req) { | |
794 | ret = -ENOMEM; | |
795 | req = orig_req; | |
796 | goto out; | |
797 | } | |
798 | ||
799 | req->r_flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE; | |
800 | ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc); | |
801 | ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid); | |
802 | ||
803 | ret = ceph_osdc_alloc_messages(req, GFP_NOFS); | |
804 | if (ret) { | |
805 | ceph_osdc_put_request(req); | |
806 | req = orig_req; | |
807 | goto out; | |
808 | } | |
809 | ||
810 | req->r_ops[0] = orig_req->r_ops[0]; | |
811 | ||
812 | req->r_mtime = aio_req->mtime; | |
813 | req->r_data_offset = req->r_ops[0].extent.offset; | |
814 | ||
815 | ceph_osdc_put_request(orig_req); | |
816 | ||
817 | req->r_callback = ceph_aio_complete_req; | |
818 | req->r_inode = inode; | |
819 | req->r_priv = aio_req; | |
820 | req->r_abort_on_full = true; | |
821 | ||
822 | ret = ceph_osdc_start_request(req->r_osdc, req, false); | |
823 | out: | |
824 | if (ret < 0) { | |
825 | req->r_result = ret; | |
826 | ceph_aio_complete_req(req); | |
827 | } | |
828 | ||
829 | ceph_put_snap_context(snapc); | |
830 | kfree(aio_work); | |
831 | } | |
832 | ||
833 | static ssize_t | |
834 | ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter, | |
835 | struct ceph_snap_context *snapc, | |
836 | struct ceph_cap_flush **pcf) | |
837 | { | |
838 | struct file *file = iocb->ki_filp; | |
839 | struct inode *inode = file_inode(file); | |
840 | struct ceph_inode_info *ci = ceph_inode(inode); | |
841 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
842 | struct ceph_vino vino; | |
843 | struct ceph_osd_request *req; | |
844 | struct page **pages; | |
845 | struct ceph_aio_request *aio_req = NULL; | |
846 | int num_pages = 0; | |
847 | int flags; | |
848 | int ret; | |
849 | struct timespec mtime = current_time(inode); | |
850 | size_t count = iov_iter_count(iter); | |
851 | loff_t pos = iocb->ki_pos; | |
852 | bool write = iov_iter_rw(iter) == WRITE; | |
853 | bool should_dirty = !write && iter_is_iovec(iter); | |
854 | ||
855 | if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP) | |
856 | return -EROFS; | |
857 | ||
858 | dout("sync_direct_%s on file %p %lld~%u snapc %p seq %lld\n", | |
859 | (write ? "write" : "read"), file, pos, (unsigned)count, | |
860 | snapc, snapc->seq); | |
861 | ||
862 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); | |
863 | if (ret < 0) | |
864 | return ret; | |
865 | ||
866 | if (write) { | |
867 | int ret2 = invalidate_inode_pages2_range(inode->i_mapping, | |
868 | pos >> PAGE_SHIFT, | |
869 | (pos + count) >> PAGE_SHIFT); | |
870 | if (ret2 < 0) | |
871 | dout("invalidate_inode_pages2_range returned %d\n", ret2); | |
872 | ||
873 | flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE; | |
874 | } else { | |
875 | flags = CEPH_OSD_FLAG_READ; | |
876 | } | |
877 | ||
878 | while (iov_iter_count(iter) > 0) { | |
879 | u64 size = dio_get_pagev_size(iter); | |
880 | size_t start = 0; | |
881 | ssize_t len; | |
882 | ||
883 | if (write) | |
884 | size = min_t(u64, size, fsc->mount_options->wsize); | |
885 | else | |
886 | size = min_t(u64, size, fsc->mount_options->rsize); | |
887 | ||
888 | vino = ceph_vino(inode); | |
889 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
890 | vino, pos, &size, 0, | |
891 | 1, | |
892 | write ? CEPH_OSD_OP_WRITE : | |
893 | CEPH_OSD_OP_READ, | |
894 | flags, snapc, | |
895 | ci->i_truncate_seq, | |
896 | ci->i_truncate_size, | |
897 | false); | |
898 | if (IS_ERR(req)) { | |
899 | ret = PTR_ERR(req); | |
900 | break; | |
901 | } | |
902 | ||
903 | len = size; | |
904 | pages = dio_get_pages_alloc(iter, len, &start, &num_pages); | |
905 | if (IS_ERR(pages)) { | |
906 | ceph_osdc_put_request(req); | |
907 | ret = PTR_ERR(pages); | |
908 | break; | |
909 | } | |
910 | ||
911 | /* | |
912 | * To simplify error handling, allow AIO when IO within i_size | |
913 | * or IO can be satisfied by single OSD request. | |
914 | */ | |
915 | if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) && | |
916 | (len == count || pos + count <= i_size_read(inode))) { | |
917 | aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL); | |
918 | if (aio_req) { | |
919 | aio_req->iocb = iocb; | |
920 | aio_req->write = write; | |
921 | aio_req->should_dirty = should_dirty; | |
922 | INIT_LIST_HEAD(&aio_req->osd_reqs); | |
923 | if (write) { | |
924 | aio_req->mtime = mtime; | |
925 | swap(aio_req->prealloc_cf, *pcf); | |
926 | } | |
927 | } | |
928 | /* ignore error */ | |
929 | } | |
930 | ||
931 | if (write) { | |
932 | /* | |
933 | * throw out any page cache pages in this range. this | |
934 | * may block. | |
935 | */ | |
936 | truncate_inode_pages_range(inode->i_mapping, pos, | |
937 | (pos+len) | (PAGE_SIZE - 1)); | |
938 | ||
939 | req->r_mtime = mtime; | |
940 | } | |
941 | ||
942 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, start, | |
943 | false, false); | |
944 | ||
945 | if (aio_req) { | |
946 | aio_req->total_len += len; | |
947 | aio_req->num_reqs++; | |
948 | atomic_inc(&aio_req->pending_reqs); | |
949 | ||
950 | req->r_callback = ceph_aio_complete_req; | |
951 | req->r_inode = inode; | |
952 | req->r_priv = aio_req; | |
953 | list_add_tail(&req->r_unsafe_item, &aio_req->osd_reqs); | |
954 | ||
955 | pos += len; | |
956 | iov_iter_advance(iter, len); | |
957 | continue; | |
958 | } | |
959 | ||
960 | ret = ceph_osdc_start_request(req->r_osdc, req, false); | |
961 | if (!ret) | |
962 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
963 | ||
964 | size = i_size_read(inode); | |
965 | if (!write) { | |
966 | if (ret == -ENOENT) | |
967 | ret = 0; | |
968 | if (ret >= 0 && ret < len && pos + ret < size) { | |
969 | int zlen = min_t(size_t, len - ret, | |
970 | size - pos - ret); | |
971 | ceph_zero_page_vector_range(start + ret, zlen, | |
972 | pages); | |
973 | ret += zlen; | |
974 | } | |
975 | if (ret >= 0) | |
976 | len = ret; | |
977 | } | |
978 | ||
979 | ceph_put_page_vector(pages, num_pages, should_dirty); | |
980 | ||
981 | ceph_osdc_put_request(req); | |
982 | if (ret < 0) | |
983 | break; | |
984 | ||
985 | pos += len; | |
986 | iov_iter_advance(iter, len); | |
987 | ||
988 | if (!write && pos >= size) | |
989 | break; | |
990 | ||
991 | if (write && pos > size) { | |
992 | if (ceph_inode_set_size(inode, pos)) | |
993 | ceph_check_caps(ceph_inode(inode), | |
994 | CHECK_CAPS_AUTHONLY, | |
995 | NULL); | |
996 | } | |
997 | } | |
998 | ||
999 | if (aio_req) { | |
1000 | LIST_HEAD(osd_reqs); | |
1001 | ||
1002 | if (aio_req->num_reqs == 0) { | |
1003 | kfree(aio_req); | |
1004 | return ret; | |
1005 | } | |
1006 | ||
1007 | ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR : | |
1008 | CEPH_CAP_FILE_RD); | |
1009 | ||
1010 | list_splice(&aio_req->osd_reqs, &osd_reqs); | |
1011 | while (!list_empty(&osd_reqs)) { | |
1012 | req = list_first_entry(&osd_reqs, | |
1013 | struct ceph_osd_request, | |
1014 | r_unsafe_item); | |
1015 | list_del_init(&req->r_unsafe_item); | |
1016 | if (ret >= 0) | |
1017 | ret = ceph_osdc_start_request(req->r_osdc, | |
1018 | req, false); | |
1019 | if (ret < 0) { | |
1020 | req->r_result = ret; | |
1021 | ceph_aio_complete_req(req); | |
1022 | } | |
1023 | } | |
1024 | return -EIOCBQUEUED; | |
1025 | } | |
1026 | ||
1027 | if (ret != -EOLDSNAPC && pos > iocb->ki_pos) { | |
1028 | ret = pos - iocb->ki_pos; | |
1029 | iocb->ki_pos = pos; | |
1030 | } | |
1031 | return ret; | |
1032 | } | |
1033 | ||
1034 | /* | |
1035 | * Synchronous write, straight from __user pointer or user pages. | |
1036 | * | |
1037 | * If write spans object boundary, just do multiple writes. (For a | |
1038 | * correct atomic write, we should e.g. take write locks on all | |
1039 | * objects, rollback on failure, etc.) | |
1040 | */ | |
1041 | static ssize_t | |
1042 | ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos, | |
1043 | struct ceph_snap_context *snapc) | |
1044 | { | |
1045 | struct file *file = iocb->ki_filp; | |
1046 | struct inode *inode = file_inode(file); | |
1047 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1048 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
1049 | struct ceph_vino vino; | |
1050 | struct ceph_osd_request *req; | |
1051 | struct page **pages; | |
1052 | u64 len; | |
1053 | int num_pages; | |
1054 | int written = 0; | |
1055 | int flags; | |
1056 | int ret; | |
1057 | bool check_caps = false; | |
1058 | struct timespec mtime = current_time(inode); | |
1059 | size_t count = iov_iter_count(from); | |
1060 | ||
1061 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) | |
1062 | return -EROFS; | |
1063 | ||
1064 | dout("sync_write on file %p %lld~%u snapc %p seq %lld\n", | |
1065 | file, pos, (unsigned)count, snapc, snapc->seq); | |
1066 | ||
1067 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); | |
1068 | if (ret < 0) | |
1069 | return ret; | |
1070 | ||
1071 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
1072 | pos >> PAGE_SHIFT, | |
1073 | (pos + count) >> PAGE_SHIFT); | |
1074 | if (ret < 0) | |
1075 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
1076 | ||
1077 | flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE; | |
1078 | ||
1079 | while ((len = iov_iter_count(from)) > 0) { | |
1080 | size_t left; | |
1081 | int n; | |
1082 | ||
1083 | vino = ceph_vino(inode); | |
1084 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1085 | vino, pos, &len, 0, 1, | |
1086 | CEPH_OSD_OP_WRITE, flags, snapc, | |
1087 | ci->i_truncate_seq, | |
1088 | ci->i_truncate_size, | |
1089 | false); | |
1090 | if (IS_ERR(req)) { | |
1091 | ret = PTR_ERR(req); | |
1092 | break; | |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * write from beginning of first page, | |
1097 | * regardless of io alignment | |
1098 | */ | |
1099 | num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1100 | ||
1101 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); | |
1102 | if (IS_ERR(pages)) { | |
1103 | ret = PTR_ERR(pages); | |
1104 | goto out; | |
1105 | } | |
1106 | ||
1107 | left = len; | |
1108 | for (n = 0; n < num_pages; n++) { | |
1109 | size_t plen = min_t(size_t, left, PAGE_SIZE); | |
1110 | ret = copy_page_from_iter(pages[n], 0, plen, from); | |
1111 | if (ret != plen) { | |
1112 | ret = -EFAULT; | |
1113 | break; | |
1114 | } | |
1115 | left -= ret; | |
1116 | } | |
1117 | ||
1118 | if (ret < 0) { | |
1119 | ceph_release_page_vector(pages, num_pages); | |
1120 | goto out; | |
1121 | } | |
1122 | ||
1123 | req->r_inode = inode; | |
1124 | ||
1125 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, | |
1126 | false, true); | |
1127 | ||
1128 | req->r_mtime = mtime; | |
1129 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); | |
1130 | if (!ret) | |
1131 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1132 | ||
1133 | out: | |
1134 | ceph_osdc_put_request(req); | |
1135 | if (ret != 0) { | |
1136 | ceph_set_error_write(ci); | |
1137 | break; | |
1138 | } | |
1139 | ||
1140 | ceph_clear_error_write(ci); | |
1141 | pos += len; | |
1142 | written += len; | |
1143 | if (pos > i_size_read(inode)) { | |
1144 | check_caps = ceph_inode_set_size(inode, pos); | |
1145 | if (check_caps) | |
1146 | ceph_check_caps(ceph_inode(inode), | |
1147 | CHECK_CAPS_AUTHONLY, | |
1148 | NULL); | |
1149 | } | |
1150 | ||
1151 | } | |
1152 | ||
1153 | if (ret != -EOLDSNAPC && written > 0) { | |
1154 | ret = written; | |
1155 | iocb->ki_pos = pos; | |
1156 | } | |
1157 | return ret; | |
1158 | } | |
1159 | ||
1160 | /* | |
1161 | * Wrap generic_file_aio_read with checks for cap bits on the inode. | |
1162 | * Atomically grab references, so that those bits are not released | |
1163 | * back to the MDS mid-read. | |
1164 | * | |
1165 | * Hmm, the sync read case isn't actually async... should it be? | |
1166 | */ | |
1167 | static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) | |
1168 | { | |
1169 | struct file *filp = iocb->ki_filp; | |
1170 | struct ceph_file_info *fi = filp->private_data; | |
1171 | size_t len = iov_iter_count(to); | |
1172 | struct inode *inode = file_inode(filp); | |
1173 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1174 | struct page *pinned_page = NULL; | |
1175 | ssize_t ret; | |
1176 | int want, got = 0; | |
1177 | int retry_op = 0, read = 0; | |
1178 | ||
1179 | again: | |
1180 | dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n", | |
1181 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode); | |
1182 | ||
1183 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1184 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
1185 | else | |
1186 | want = CEPH_CAP_FILE_CACHE; | |
1187 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page); | |
1188 | if (ret < 0) | |
1189 | return ret; | |
1190 | ||
1191 | if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 || | |
1192 | (iocb->ki_flags & IOCB_DIRECT) || | |
1193 | (fi->flags & CEPH_F_SYNC)) { | |
1194 | ||
1195 | dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n", | |
1196 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, | |
1197 | ceph_cap_string(got)); | |
1198 | ||
1199 | if (ci->i_inline_version == CEPH_INLINE_NONE) { | |
1200 | if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) { | |
1201 | ret = ceph_direct_read_write(iocb, to, | |
1202 | NULL, NULL); | |
1203 | if (ret >= 0 && ret < len) | |
1204 | retry_op = CHECK_EOF; | |
1205 | } else { | |
1206 | ret = ceph_sync_read(iocb, to, &retry_op); | |
1207 | } | |
1208 | } else { | |
1209 | retry_op = READ_INLINE; | |
1210 | } | |
1211 | } else { | |
1212 | CEPH_DEFINE_RW_CONTEXT(rw_ctx, got); | |
1213 | dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n", | |
1214 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, | |
1215 | ceph_cap_string(got)); | |
1216 | ceph_add_rw_context(fi, &rw_ctx); | |
1217 | ret = generic_file_read_iter(iocb, to); | |
1218 | ceph_del_rw_context(fi, &rw_ctx); | |
1219 | } | |
1220 | dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", | |
1221 | inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); | |
1222 | if (pinned_page) { | |
1223 | put_page(pinned_page); | |
1224 | pinned_page = NULL; | |
1225 | } | |
1226 | ceph_put_cap_refs(ci, got); | |
1227 | if (retry_op > HAVE_RETRIED && ret >= 0) { | |
1228 | int statret; | |
1229 | struct page *page = NULL; | |
1230 | loff_t i_size; | |
1231 | if (retry_op == READ_INLINE) { | |
1232 | page = __page_cache_alloc(GFP_KERNEL); | |
1233 | if (!page) | |
1234 | return -ENOMEM; | |
1235 | } | |
1236 | ||
1237 | statret = __ceph_do_getattr(inode, page, | |
1238 | CEPH_STAT_CAP_INLINE_DATA, !!page); | |
1239 | if (statret < 0) { | |
1240 | if (page) | |
1241 | __free_page(page); | |
1242 | if (statret == -ENODATA) { | |
1243 | BUG_ON(retry_op != READ_INLINE); | |
1244 | goto again; | |
1245 | } | |
1246 | return statret; | |
1247 | } | |
1248 | ||
1249 | i_size = i_size_read(inode); | |
1250 | if (retry_op == READ_INLINE) { | |
1251 | BUG_ON(ret > 0 || read > 0); | |
1252 | if (iocb->ki_pos < i_size && | |
1253 | iocb->ki_pos < PAGE_SIZE) { | |
1254 | loff_t end = min_t(loff_t, i_size, | |
1255 | iocb->ki_pos + len); | |
1256 | end = min_t(loff_t, end, PAGE_SIZE); | |
1257 | if (statret < end) | |
1258 | zero_user_segment(page, statret, end); | |
1259 | ret = copy_page_to_iter(page, | |
1260 | iocb->ki_pos & ~PAGE_MASK, | |
1261 | end - iocb->ki_pos, to); | |
1262 | iocb->ki_pos += ret; | |
1263 | read += ret; | |
1264 | } | |
1265 | if (iocb->ki_pos < i_size && read < len) { | |
1266 | size_t zlen = min_t(size_t, len - read, | |
1267 | i_size - iocb->ki_pos); | |
1268 | ret = iov_iter_zero(zlen, to); | |
1269 | iocb->ki_pos += ret; | |
1270 | read += ret; | |
1271 | } | |
1272 | __free_pages(page, 0); | |
1273 | return read; | |
1274 | } | |
1275 | ||
1276 | /* hit EOF or hole? */ | |
1277 | if (retry_op == CHECK_EOF && iocb->ki_pos < i_size && | |
1278 | ret < len) { | |
1279 | dout("sync_read hit hole, ppos %lld < size %lld" | |
1280 | ", reading more\n", iocb->ki_pos, i_size); | |
1281 | ||
1282 | read += ret; | |
1283 | len -= ret; | |
1284 | retry_op = HAVE_RETRIED; | |
1285 | goto again; | |
1286 | } | |
1287 | } | |
1288 | ||
1289 | if (ret >= 0) | |
1290 | ret += read; | |
1291 | ||
1292 | return ret; | |
1293 | } | |
1294 | ||
1295 | /* | |
1296 | * Take cap references to avoid releasing caps to MDS mid-write. | |
1297 | * | |
1298 | * If we are synchronous, and write with an old snap context, the OSD | |
1299 | * may return EOLDSNAPC. In that case, retry the write.. _after_ | |
1300 | * dropping our cap refs and allowing the pending snap to logically | |
1301 | * complete _before_ this write occurs. | |
1302 | * | |
1303 | * If we are near ENOSPC, write synchronously. | |
1304 | */ | |
1305 | static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from) | |
1306 | { | |
1307 | struct file *file = iocb->ki_filp; | |
1308 | struct ceph_file_info *fi = file->private_data; | |
1309 | struct inode *inode = file_inode(file); | |
1310 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1311 | struct ceph_osd_client *osdc = | |
1312 | &ceph_sb_to_client(inode->i_sb)->client->osdc; | |
1313 | struct ceph_cap_flush *prealloc_cf; | |
1314 | ssize_t count, written = 0; | |
1315 | int err, want, got; | |
1316 | loff_t pos; | |
1317 | ||
1318 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
1319 | return -EROFS; | |
1320 | ||
1321 | prealloc_cf = ceph_alloc_cap_flush(); | |
1322 | if (!prealloc_cf) | |
1323 | return -ENOMEM; | |
1324 | ||
1325 | retry_snap: | |
1326 | inode_lock(inode); | |
1327 | ||
1328 | /* We can write back this queue in page reclaim */ | |
1329 | current->backing_dev_info = inode_to_bdi(inode); | |
1330 | ||
1331 | if (iocb->ki_flags & IOCB_APPEND) { | |
1332 | err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); | |
1333 | if (err < 0) | |
1334 | goto out; | |
1335 | } | |
1336 | ||
1337 | err = generic_write_checks(iocb, from); | |
1338 | if (err <= 0) | |
1339 | goto out; | |
1340 | ||
1341 | pos = iocb->ki_pos; | |
1342 | count = iov_iter_count(from); | |
1343 | if (ceph_quota_is_max_bytes_exceeded(inode, pos + count)) { | |
1344 | err = -EDQUOT; | |
1345 | goto out; | |
1346 | } | |
1347 | ||
1348 | err = file_remove_privs(file); | |
1349 | if (err) | |
1350 | goto out; | |
1351 | ||
1352 | err = file_update_time(file); | |
1353 | if (err) | |
1354 | goto out; | |
1355 | ||
1356 | if (ci->i_inline_version != CEPH_INLINE_NONE) { | |
1357 | err = ceph_uninline_data(file, NULL); | |
1358 | if (err < 0) | |
1359 | goto out; | |
1360 | } | |
1361 | ||
1362 | /* FIXME: not complete since it doesn't account for being at quota */ | |
1363 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL)) { | |
1364 | err = -ENOSPC; | |
1365 | goto out; | |
1366 | } | |
1367 | ||
1368 | dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n", | |
1369 | inode, ceph_vinop(inode), pos, count, i_size_read(inode)); | |
1370 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1371 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1372 | else | |
1373 | want = CEPH_CAP_FILE_BUFFER; | |
1374 | got = 0; | |
1375 | err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count, | |
1376 | &got, NULL); | |
1377 | if (err < 0) | |
1378 | goto out; | |
1379 | ||
1380 | dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n", | |
1381 | inode, ceph_vinop(inode), pos, count, ceph_cap_string(got)); | |
1382 | ||
1383 | if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || | |
1384 | (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC) || | |
1385 | (ci->i_ceph_flags & CEPH_I_ERROR_WRITE)) { | |
1386 | struct ceph_snap_context *snapc; | |
1387 | struct iov_iter data; | |
1388 | inode_unlock(inode); | |
1389 | ||
1390 | spin_lock(&ci->i_ceph_lock); | |
1391 | if (__ceph_have_pending_cap_snap(ci)) { | |
1392 | struct ceph_cap_snap *capsnap = | |
1393 | list_last_entry(&ci->i_cap_snaps, | |
1394 | struct ceph_cap_snap, | |
1395 | ci_item); | |
1396 | snapc = ceph_get_snap_context(capsnap->context); | |
1397 | } else { | |
1398 | BUG_ON(!ci->i_head_snapc); | |
1399 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
1400 | } | |
1401 | spin_unlock(&ci->i_ceph_lock); | |
1402 | ||
1403 | /* we might need to revert back to that point */ | |
1404 | data = *from; | |
1405 | if (iocb->ki_flags & IOCB_DIRECT) | |
1406 | written = ceph_direct_read_write(iocb, &data, snapc, | |
1407 | &prealloc_cf); | |
1408 | else | |
1409 | written = ceph_sync_write(iocb, &data, pos, snapc); | |
1410 | if (written > 0) | |
1411 | iov_iter_advance(from, written); | |
1412 | ceph_put_snap_context(snapc); | |
1413 | } else { | |
1414 | /* | |
1415 | * No need to acquire the i_truncate_mutex. Because | |
1416 | * the MDS revokes Fwb caps before sending truncate | |
1417 | * message to us. We can't get Fwb cap while there | |
1418 | * are pending vmtruncate. So write and vmtruncate | |
1419 | * can not run at the same time | |
1420 | */ | |
1421 | written = generic_perform_write(file, from, pos); | |
1422 | if (likely(written >= 0)) | |
1423 | iocb->ki_pos = pos + written; | |
1424 | inode_unlock(inode); | |
1425 | } | |
1426 | ||
1427 | if (written >= 0) { | |
1428 | int dirty; | |
1429 | ||
1430 | spin_lock(&ci->i_ceph_lock); | |
1431 | ci->i_inline_version = CEPH_INLINE_NONE; | |
1432 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, | |
1433 | &prealloc_cf); | |
1434 | spin_unlock(&ci->i_ceph_lock); | |
1435 | if (dirty) | |
1436 | __mark_inode_dirty(inode, dirty); | |
1437 | if (ceph_quota_is_max_bytes_approaching(inode, iocb->ki_pos)) | |
1438 | ceph_check_caps(ci, CHECK_CAPS_NODELAY, NULL); | |
1439 | } | |
1440 | ||
1441 | dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n", | |
1442 | inode, ceph_vinop(inode), pos, (unsigned)count, | |
1443 | ceph_cap_string(got)); | |
1444 | ceph_put_cap_refs(ci, got); | |
1445 | ||
1446 | if (written == -EOLDSNAPC) { | |
1447 | dout("aio_write %p %llx.%llx %llu~%u" "got EOLDSNAPC, retrying\n", | |
1448 | inode, ceph_vinop(inode), pos, (unsigned)count); | |
1449 | goto retry_snap; | |
1450 | } | |
1451 | ||
1452 | if (written >= 0) { | |
1453 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_NEARFULL)) | |
1454 | iocb->ki_flags |= IOCB_DSYNC; | |
1455 | written = generic_write_sync(iocb, written); | |
1456 | } | |
1457 | ||
1458 | goto out_unlocked; | |
1459 | ||
1460 | out: | |
1461 | inode_unlock(inode); | |
1462 | out_unlocked: | |
1463 | ceph_free_cap_flush(prealloc_cf); | |
1464 | current->backing_dev_info = NULL; | |
1465 | return written ? written : err; | |
1466 | } | |
1467 | ||
1468 | /* | |
1469 | * llseek. be sure to verify file size on SEEK_END. | |
1470 | */ | |
1471 | static loff_t ceph_llseek(struct file *file, loff_t offset, int whence) | |
1472 | { | |
1473 | struct inode *inode = file->f_mapping->host; | |
1474 | loff_t i_size; | |
1475 | loff_t ret; | |
1476 | ||
1477 | inode_lock(inode); | |
1478 | ||
1479 | if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { | |
1480 | ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); | |
1481 | if (ret < 0) | |
1482 | goto out; | |
1483 | } | |
1484 | ||
1485 | i_size = i_size_read(inode); | |
1486 | switch (whence) { | |
1487 | case SEEK_END: | |
1488 | offset += i_size; | |
1489 | break; | |
1490 | case SEEK_CUR: | |
1491 | /* | |
1492 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
1493 | * position-querying operation. Avoid rewriting the "same" | |
1494 | * f_pos value back to the file because a concurrent read(), | |
1495 | * write() or lseek() might have altered it | |
1496 | */ | |
1497 | if (offset == 0) { | |
1498 | ret = file->f_pos; | |
1499 | goto out; | |
1500 | } | |
1501 | offset += file->f_pos; | |
1502 | break; | |
1503 | case SEEK_DATA: | |
1504 | if (offset < 0 || offset >= i_size) { | |
1505 | ret = -ENXIO; | |
1506 | goto out; | |
1507 | } | |
1508 | break; | |
1509 | case SEEK_HOLE: | |
1510 | if (offset < 0 || offset >= i_size) { | |
1511 | ret = -ENXIO; | |
1512 | goto out; | |
1513 | } | |
1514 | offset = i_size; | |
1515 | break; | |
1516 | } | |
1517 | ||
1518 | ret = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); | |
1519 | ||
1520 | out: | |
1521 | inode_unlock(inode); | |
1522 | return ret; | |
1523 | } | |
1524 | ||
1525 | static inline void ceph_zero_partial_page( | |
1526 | struct inode *inode, loff_t offset, unsigned size) | |
1527 | { | |
1528 | struct page *page; | |
1529 | pgoff_t index = offset >> PAGE_SHIFT; | |
1530 | ||
1531 | page = find_lock_page(inode->i_mapping, index); | |
1532 | if (page) { | |
1533 | wait_on_page_writeback(page); | |
1534 | zero_user(page, offset & (PAGE_SIZE - 1), size); | |
1535 | unlock_page(page); | |
1536 | put_page(page); | |
1537 | } | |
1538 | } | |
1539 | ||
1540 | static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset, | |
1541 | loff_t length) | |
1542 | { | |
1543 | loff_t nearly = round_up(offset, PAGE_SIZE); | |
1544 | if (offset < nearly) { | |
1545 | loff_t size = nearly - offset; | |
1546 | if (length < size) | |
1547 | size = length; | |
1548 | ceph_zero_partial_page(inode, offset, size); | |
1549 | offset += size; | |
1550 | length -= size; | |
1551 | } | |
1552 | if (length >= PAGE_SIZE) { | |
1553 | loff_t size = round_down(length, PAGE_SIZE); | |
1554 | truncate_pagecache_range(inode, offset, offset + size - 1); | |
1555 | offset += size; | |
1556 | length -= size; | |
1557 | } | |
1558 | if (length) | |
1559 | ceph_zero_partial_page(inode, offset, length); | |
1560 | } | |
1561 | ||
1562 | static int ceph_zero_partial_object(struct inode *inode, | |
1563 | loff_t offset, loff_t *length) | |
1564 | { | |
1565 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1566 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
1567 | struct ceph_osd_request *req; | |
1568 | int ret = 0; | |
1569 | loff_t zero = 0; | |
1570 | int op; | |
1571 | ||
1572 | if (!length) { | |
1573 | op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE; | |
1574 | length = &zero; | |
1575 | } else { | |
1576 | op = CEPH_OSD_OP_ZERO; | |
1577 | } | |
1578 | ||
1579 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1580 | ceph_vino(inode), | |
1581 | offset, length, | |
1582 | 0, 1, op, | |
1583 | CEPH_OSD_FLAG_WRITE, | |
1584 | NULL, 0, 0, false); | |
1585 | if (IS_ERR(req)) { | |
1586 | ret = PTR_ERR(req); | |
1587 | goto out; | |
1588 | } | |
1589 | ||
1590 | req->r_mtime = inode->i_mtime; | |
1591 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); | |
1592 | if (!ret) { | |
1593 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1594 | if (ret == -ENOENT) | |
1595 | ret = 0; | |
1596 | } | |
1597 | ceph_osdc_put_request(req); | |
1598 | ||
1599 | out: | |
1600 | return ret; | |
1601 | } | |
1602 | ||
1603 | static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length) | |
1604 | { | |
1605 | int ret = 0; | |
1606 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1607 | s32 stripe_unit = ci->i_layout.stripe_unit; | |
1608 | s32 stripe_count = ci->i_layout.stripe_count; | |
1609 | s32 object_size = ci->i_layout.object_size; | |
1610 | u64 object_set_size = object_size * stripe_count; | |
1611 | u64 nearly, t; | |
1612 | ||
1613 | /* round offset up to next period boundary */ | |
1614 | nearly = offset + object_set_size - 1; | |
1615 | t = nearly; | |
1616 | nearly -= do_div(t, object_set_size); | |
1617 | ||
1618 | while (length && offset < nearly) { | |
1619 | loff_t size = length; | |
1620 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1621 | if (ret < 0) | |
1622 | return ret; | |
1623 | offset += size; | |
1624 | length -= size; | |
1625 | } | |
1626 | while (length >= object_set_size) { | |
1627 | int i; | |
1628 | loff_t pos = offset; | |
1629 | for (i = 0; i < stripe_count; ++i) { | |
1630 | ret = ceph_zero_partial_object(inode, pos, NULL); | |
1631 | if (ret < 0) | |
1632 | return ret; | |
1633 | pos += stripe_unit; | |
1634 | } | |
1635 | offset += object_set_size; | |
1636 | length -= object_set_size; | |
1637 | } | |
1638 | while (length) { | |
1639 | loff_t size = length; | |
1640 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1641 | if (ret < 0) | |
1642 | return ret; | |
1643 | offset += size; | |
1644 | length -= size; | |
1645 | } | |
1646 | return ret; | |
1647 | } | |
1648 | ||
1649 | static long ceph_fallocate(struct file *file, int mode, | |
1650 | loff_t offset, loff_t length) | |
1651 | { | |
1652 | struct ceph_file_info *fi = file->private_data; | |
1653 | struct inode *inode = file_inode(file); | |
1654 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1655 | struct ceph_osd_client *osdc = | |
1656 | &ceph_inode_to_client(inode)->client->osdc; | |
1657 | struct ceph_cap_flush *prealloc_cf; | |
1658 | int want, got = 0; | |
1659 | int dirty; | |
1660 | int ret = 0; | |
1661 | loff_t endoff = 0; | |
1662 | loff_t size; | |
1663 | ||
1664 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
1665 | return -EOPNOTSUPP; | |
1666 | ||
1667 | if (!S_ISREG(inode->i_mode)) | |
1668 | return -EOPNOTSUPP; | |
1669 | ||
1670 | prealloc_cf = ceph_alloc_cap_flush(); | |
1671 | if (!prealloc_cf) | |
1672 | return -ENOMEM; | |
1673 | ||
1674 | inode_lock(inode); | |
1675 | ||
1676 | if (ceph_snap(inode) != CEPH_NOSNAP) { | |
1677 | ret = -EROFS; | |
1678 | goto unlock; | |
1679 | } | |
1680 | ||
1681 | if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) && | |
1682 | ceph_quota_is_max_bytes_exceeded(inode, offset + length)) { | |
1683 | ret = -EDQUOT; | |
1684 | goto unlock; | |
1685 | } | |
1686 | ||
1687 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) && | |
1688 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
1689 | ret = -ENOSPC; | |
1690 | goto unlock; | |
1691 | } | |
1692 | ||
1693 | if (ci->i_inline_version != CEPH_INLINE_NONE) { | |
1694 | ret = ceph_uninline_data(file, NULL); | |
1695 | if (ret < 0) | |
1696 | goto unlock; | |
1697 | } | |
1698 | ||
1699 | size = i_size_read(inode); | |
1700 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { | |
1701 | endoff = offset + length; | |
1702 | ret = inode_newsize_ok(inode, endoff); | |
1703 | if (ret) | |
1704 | goto unlock; | |
1705 | } | |
1706 | ||
1707 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1708 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1709 | else | |
1710 | want = CEPH_CAP_FILE_BUFFER; | |
1711 | ||
1712 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL); | |
1713 | if (ret < 0) | |
1714 | goto unlock; | |
1715 | ||
1716 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1717 | if (offset < size) | |
1718 | ceph_zero_pagecache_range(inode, offset, length); | |
1719 | ret = ceph_zero_objects(inode, offset, length); | |
1720 | } else if (endoff > size) { | |
1721 | truncate_pagecache_range(inode, size, -1); | |
1722 | if (ceph_inode_set_size(inode, endoff)) | |
1723 | ceph_check_caps(ceph_inode(inode), | |
1724 | CHECK_CAPS_AUTHONLY, NULL); | |
1725 | } | |
1726 | ||
1727 | if (!ret) { | |
1728 | spin_lock(&ci->i_ceph_lock); | |
1729 | ci->i_inline_version = CEPH_INLINE_NONE; | |
1730 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, | |
1731 | &prealloc_cf); | |
1732 | spin_unlock(&ci->i_ceph_lock); | |
1733 | if (dirty) | |
1734 | __mark_inode_dirty(inode, dirty); | |
1735 | if ((endoff > size) && | |
1736 | ceph_quota_is_max_bytes_approaching(inode, endoff)) | |
1737 | ceph_check_caps(ci, CHECK_CAPS_NODELAY, NULL); | |
1738 | } | |
1739 | ||
1740 | ceph_put_cap_refs(ci, got); | |
1741 | unlock: | |
1742 | inode_unlock(inode); | |
1743 | ceph_free_cap_flush(prealloc_cf); | |
1744 | return ret; | |
1745 | } | |
1746 | ||
1747 | const struct file_operations ceph_file_fops = { | |
1748 | .open = ceph_open, | |
1749 | .release = ceph_release, | |
1750 | .llseek = ceph_llseek, | |
1751 | .read_iter = ceph_read_iter, | |
1752 | .write_iter = ceph_write_iter, | |
1753 | .mmap = ceph_mmap, | |
1754 | .fsync = ceph_fsync, | |
1755 | .lock = ceph_lock, | |
1756 | .flock = ceph_flock, | |
1757 | .splice_read = generic_file_splice_read, | |
1758 | .splice_write = iter_file_splice_write, | |
1759 | .unlocked_ioctl = ceph_ioctl, | |
1760 | .compat_ioctl = ceph_ioctl, | |
1761 | .fallocate = ceph_fallocate, | |
1762 | }; | |
1763 |