]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
3d14c5d2 | 2 | #include <linux/ceph/ceph_debug.h> |
124e68e7 | 3 | |
3d14c5d2 | 4 | #include <linux/module.h> |
124e68e7 | 5 | #include <linux/sched.h> |
5a0e3ad6 | 6 | #include <linux/slab.h> |
124e68e7 | 7 | #include <linux/file.h> |
5ef50c3b | 8 | #include <linux/mount.h> |
124e68e7 SW |
9 | #include <linux/namei.h> |
10 | #include <linux/writeback.h> | |
ad7a60de | 11 | #include <linux/falloc.h> |
124e68e7 SW |
12 | |
13 | #include "super.h" | |
14 | #include "mds_client.h" | |
99ccbd22 | 15 | #include "cache.h" |
124e68e7 | 16 | |
f775ff7d AG |
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 | ||
124e68e7 SW |
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 | ||
b5b98989 ZC |
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); | |
752ade68 MH |
110 | pages = kvmalloc(sizeof(*pages) * npages, GFP_KERNEL); |
111 | if (!pages) | |
112 | return ERR_PTR(-ENOMEM); | |
b5b98989 ZC |
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 | } | |
124e68e7 SW |
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 | { | |
3d14c5d2 YS |
143 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
144 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
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); | |
f775ff7d | 156 | req->r_args.open.flags = ceph_flags_sys2wire(flags); |
124e68e7 | 157 | req->r_args.open.mode = cpu_to_le32(create_mode); |
124e68e7 SW |
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: | |
46b59b2b YZ |
173 | ceph_fscache_register_inode_cookie(inode); |
174 | ceph_fscache_file_set_cookie(inode, file); | |
124e68e7 SW |
175 | case S_IFDIR: |
176 | dout("init_file %p %p 0%o (regular)\n", inode, file, | |
177 | inode->i_mode); | |
99ec2697 | 178 | cf = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL); |
d37b1d99 | 179 | if (!cf) { |
124e68e7 SW |
180 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ |
181 | return -ENOMEM; | |
182 | } | |
183 | cf->fmode = fmode; | |
3bd4d543 YZ |
184 | |
185 | spin_lock_init(&cf->rw_contexts_lock); | |
186 | INIT_LIST_HEAD(&cf->rw_contexts); | |
187 | ||
124e68e7 | 188 | cf->next_offset = 2; |
fdd4e158 | 189 | cf->readdir_cache_idx = -1; |
124e68e7 SW |
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 | ||
77310320 YZ |
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) && | |
8242c9f3 | 229 | (!(wanted & CEPH_CAP_ANY_WR) || ci->i_auth_cap)) { |
77310320 YZ |
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 | ||
124e68e7 | 269 | /* |
124e68e7 SW |
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); | |
3d14c5d2 YS |
278 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
279 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
280 | struct ceph_mds_request *req; |
281 | struct ceph_file_info *cf = file->private_data; | |
124e68e7 SW |
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) { | |
be655596 | 306 | spin_lock(&ci->i_ceph_lock); |
124e68e7 | 307 | __ceph_get_fmode(ci, fmode); |
be655596 | 308 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
309 | return ceph_init_file(inode, file, fmode); |
310 | } | |
311 | ||
312 | /* | |
7421ab80 SW |
313 | * No need to block if we have caps on the auth MDS (for |
314 | * write) or any MDS (for read). Update wanted set | |
124e68e7 SW |
315 | * asynchronously. |
316 | */ | |
be655596 | 317 | spin_lock(&ci->i_ceph_lock); |
7421ab80 SW |
318 | if (__ceph_is_any_real_caps(ci) && |
319 | (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) { | |
c1944fed | 320 | int mds_wanted = __ceph_caps_mds_wanted(ci, true); |
124e68e7 SW |
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); | |
be655596 | 327 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
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); | |
be655596 | 339 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
340 | return ceph_init_file(inode, file, fmode); |
341 | } | |
99ccbd22 | 342 | |
be655596 | 343 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
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 | } | |
70b666c3 SW |
351 | req->r_inode = inode; |
352 | ihold(inode); | |
99ccbd22 | 353 | |
124e68e7 | 354 | req->r_num_caps = 1; |
e36d571d | 355 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
124e68e7 SW |
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 | /* | |
5ef50c3b SW |
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. | |
124e68e7 | 368 | */ |
5ef50c3b | 369 | int ceph_atomic_open(struct inode *dir, struct dentry *dentry, |
30d90494 | 370 | struct file *file, unsigned flags, umode_t mode, |
d9585277 | 371 | int *opened) |
124e68e7 | 372 | { |
3d14c5d2 YS |
373 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
374 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 | 375 | struct ceph_mds_request *req; |
5ef50c3b | 376 | struct dentry *dn; |
b1ee94aa | 377 | struct ceph_acls_info acls = {}; |
315f2408 | 378 | int mask; |
124e68e7 | 379 | int err; |
124e68e7 | 380 | |
a455589f AV |
381 | dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n", |
382 | dir, dentry, dentry, | |
5ef50c3b SW |
383 | d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode); |
384 | ||
385 | if (dentry->d_name.len > NAME_MAX) | |
386 | return -ENAMETOOLONG; | |
387 | ||
b1ee94aa YZ |
388 | if (flags & O_CREAT) { |
389 | err = ceph_pre_init_acls(dir, &mode, &acls); | |
390 | if (err < 0) | |
391 | return err; | |
392 | } | |
393 | ||
124e68e7 SW |
394 | /* do the open */ |
395 | req = prepare_open_request(dir->i_sb, flags, mode); | |
b1ee94aa YZ |
396 | if (IS_ERR(req)) { |
397 | err = PTR_ERR(req); | |
398 | goto out_acl; | |
399 | } | |
124e68e7 SW |
400 | req->r_dentry = dget(dentry); |
401 | req->r_num_caps = 2; | |
402 | if (flags & O_CREAT) { | |
403 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
404 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
b1ee94aa YZ |
405 | if (acls.pagelist) { |
406 | req->r_pagelist = acls.pagelist; | |
407 | acls.pagelist = NULL; | |
408 | } | |
124e68e7 | 409 | } |
315f2408 YZ |
410 | |
411 | mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED; | |
412 | if (ceph_security_xattr_wanted(dir)) | |
413 | mask |= CEPH_CAP_XATTR_SHARED; | |
414 | req->r_args.open.mask = cpu_to_le32(mask); | |
415 | ||
3dd69aab JL |
416 | req->r_parent = dir; |
417 | set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); | |
acda7657 SW |
418 | err = ceph_mdsc_do_request(mdsc, |
419 | (flags & (O_CREAT|O_TRUNC)) ? dir : NULL, | |
420 | req); | |
bf91c315 | 421 | err = ceph_handle_snapdir(req, dentry, err); |
79aec984 | 422 | if (err) |
b1ee94aa | 423 | goto out_req; |
79aec984 | 424 | |
a43137f7 | 425 | if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry) |
124e68e7 | 426 | err = ceph_handle_notrace_create(dir, dentry); |
2d83bde9 | 427 | |
00699ad8 | 428 | if (d_in_lookup(dentry)) { |
5ef50c3b SW |
429 | dn = ceph_finish_lookup(req, dentry, err); |
430 | if (IS_ERR(dn)) | |
431 | err = PTR_ERR(dn); | |
432 | } else { | |
433 | /* we were given a hashed negative dentry */ | |
434 | dn = NULL; | |
435 | } | |
436 | if (err) | |
b1ee94aa | 437 | goto out_req; |
2b0143b5 | 438 | if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) { |
5ef50c3b SW |
439 | /* make vfs retry on splice, ENOENT, or symlink */ |
440 | dout("atomic_open finish_no_open on dn %p\n", dn); | |
441 | err = finish_no_open(file, dn); | |
442 | } else { | |
443 | dout("atomic_open finish_open on dn %p\n", dn); | |
6e8575fa | 444 | if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) { |
2b0143b5 | 445 | ceph_init_inode_acls(d_inode(dentry), &acls); |
6e8575fa SL |
446 | *opened |= FILE_CREATED; |
447 | } | |
5ef50c3b SW |
448 | err = finish_open(file, dentry, ceph_open, opened); |
449 | } | |
b1ee94aa | 450 | out_req: |
ab866549 YZ |
451 | if (!req->r_err && req->r_target_inode) |
452 | ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode); | |
5ef50c3b | 453 | ceph_mdsc_put_request(req); |
b1ee94aa YZ |
454 | out_acl: |
455 | ceph_release_acls_info(&acls); | |
5ef50c3b | 456 | dout("atomic_open result=%d\n", err); |
d9585277 | 457 | return err; |
124e68e7 SW |
458 | } |
459 | ||
460 | int ceph_release(struct inode *inode, struct file *file) | |
461 | { | |
462 | struct ceph_inode_info *ci = ceph_inode(inode); | |
463 | struct ceph_file_info *cf = file->private_data; | |
464 | ||
465 | dout("release inode %p file %p\n", inode, file); | |
466 | ceph_put_fmode(ci, cf->fmode); | |
467 | if (cf->last_readdir) | |
468 | ceph_mdsc_put_request(cf->last_readdir); | |
469 | kfree(cf->last_name); | |
470 | kfree(cf->dir_info); | |
3bd4d543 | 471 | WARN_ON(!list_empty(&cf->rw_contexts)); |
124e68e7 | 472 | kmem_cache_free(ceph_file_cachep, cf); |
195d3ce2 SW |
473 | |
474 | /* wake up anyone waiting for caps on this inode */ | |
03066f23 | 475 | wake_up_all(&ci->i_cap_wq); |
124e68e7 SW |
476 | return 0; |
477 | } | |
478 | ||
83701246 | 479 | enum { |
c8fe9b17 YZ |
480 | HAVE_RETRIED = 1, |
481 | CHECK_EOF = 2, | |
482 | READ_INLINE = 3, | |
83701246 YZ |
483 | }; |
484 | ||
124e68e7 SW |
485 | /* |
486 | * Read a range of bytes striped over one or more objects. Iterate over | |
487 | * objects we stripe over. (That's not atomic, but good enough for now.) | |
488 | * | |
489 | * If we get a short result from the OSD, check against i_size; we need to | |
490 | * only return a short read to the caller if we hit EOF. | |
491 | */ | |
492 | static int striped_read(struct inode *inode, | |
7ce469a5 | 493 | u64 pos, u64 len, |
6a026589 | 494 | struct page **pages, int num_pages, |
7ce469a5 | 495 | int page_align, int *checkeof) |
124e68e7 | 496 | { |
3d14c5d2 | 497 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
124e68e7 | 498 | struct ceph_inode_info *ci = ceph_inode(inode); |
7ce469a5 | 499 | u64 this_len; |
99c88e69 | 500 | loff_t i_size; |
7ce469a5 YZ |
501 | int page_idx; |
502 | int ret, read = 0; | |
124e68e7 SW |
503 | bool hit_stripe, was_short; |
504 | ||
505 | /* | |
506 | * we may need to do multiple reads. not atomic, unfortunately. | |
507 | */ | |
124e68e7 | 508 | more: |
7ce469a5 YZ |
509 | this_len = len; |
510 | page_idx = (page_align + read) >> PAGE_SHIFT; | |
3d14c5d2 | 511 | ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), |
124e68e7 | 512 | &ci->i_layout, pos, &this_len, |
7ce469a5 YZ |
513 | ci->i_truncate_seq, ci->i_truncate_size, |
514 | pages + page_idx, num_pages - page_idx, | |
515 | ((page_align + read) & ~PAGE_MASK)); | |
124e68e7 SW |
516 | if (ret == -ENOENT) |
517 | ret = 0; | |
7ce469a5 | 518 | hit_stripe = this_len < len; |
0e98728f | 519 | was_short = ret >= 0 && ret < this_len; |
7ce469a5 | 520 | dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, len, read, |
124e68e7 SW |
521 | ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : ""); |
522 | ||
99c88e69 | 523 | i_size = i_size_read(inode); |
02ae66d8 | 524 | if (ret >= 0) { |
99c88e69 YZ |
525 | if (was_short && (pos + ret < i_size)) { |
526 | int zlen = min(this_len - ret, i_size - pos - ret); | |
7ce469a5 | 527 | int zoff = page_align + read + ret; |
02ae66d8 | 528 | dout(" zero gap %llu to %llu\n", |
7ce469a5 | 529 | pos + ret, pos + ret + zlen); |
1487a688 YZ |
530 | ceph_zero_page_vector_range(zoff, zlen, pages); |
531 | ret += zlen; | |
124e68e7 | 532 | } |
02ae66d8 | 533 | |
7ce469a5 | 534 | read += ret; |
124e68e7 | 535 | pos += ret; |
7ce469a5 | 536 | len -= ret; |
124e68e7 | 537 | |
02ae66d8 | 538 | /* hit stripe and need continue*/ |
7ce469a5 | 539 | if (len && hit_stripe && pos < i_size) |
124e68e7 SW |
540 | goto more; |
541 | } | |
542 | ||
ee7289bf | 543 | if (read > 0) { |
02ae66d8 | 544 | ret = read; |
c3cd6283 | 545 | /* did we bounce off eof? */ |
7ce469a5 | 546 | if (pos + len > i_size) |
83701246 | 547 | *checkeof = CHECK_EOF; |
124e68e7 SW |
548 | } |
549 | ||
124e68e7 SW |
550 | dout("striped_read returns %d\n", ret); |
551 | return ret; | |
552 | } | |
553 | ||
554 | /* | |
555 | * Completely synchronous read and write methods. Direct from __user | |
556 | * buffer to osd, or directly to user pages (if O_DIRECT). | |
557 | * | |
558 | * If the read spans object boundary, just do multiple reads. | |
559 | */ | |
7ce469a5 YZ |
560 | static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to, |
561 | int *checkeof) | |
124e68e7 | 562 | { |
8eb4efb0 | 563 | struct file *file = iocb->ki_filp; |
496ad9aa | 564 | struct inode *inode = file_inode(file); |
124e68e7 | 565 | struct page **pages; |
8eb4efb0 | 566 | u64 off = iocb->ki_pos; |
7ce469a5 YZ |
567 | int num_pages; |
568 | ssize_t ret; | |
569 | size_t len = iov_iter_count(to); | |
124e68e7 | 570 | |
1c0a9c2d | 571 | dout("sync_read on file %p %llu~%u %s\n", file, off, (unsigned)len, |
124e68e7 | 572 | (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); |
d0d0db22 YZ |
573 | |
574 | if (!len) | |
575 | return 0; | |
e98b6fed SW |
576 | /* |
577 | * flush any page cache pages in this range. this | |
578 | * will make concurrent normal and sync io slow, | |
579 | * but it will at least behave sensibly when they are | |
580 | * in sequence. | |
581 | */ | |
8eb4efb0 | 582 | ret = filemap_write_and_wait_range(inode->i_mapping, off, |
583 | off + len); | |
29065a51 | 584 | if (ret < 0) |
8eb4efb0 | 585 | return ret; |
29065a51 | 586 | |
7ce469a5 YZ |
587 | if (unlikely(to->type & ITER_PIPE)) { |
588 | size_t page_off; | |
589 | ret = iov_iter_get_pages_alloc(to, &pages, len, | |
590 | &page_off); | |
591 | if (ret <= 0) | |
592 | return -ENOMEM; | |
593 | num_pages = DIV_ROUND_UP(ret + page_off, PAGE_SIZE); | |
594 | ||
595 | ret = striped_read(inode, off, ret, pages, num_pages, | |
596 | page_off, checkeof); | |
597 | if (ret > 0) { | |
598 | iov_iter_advance(to, ret); | |
599 | off += ret; | |
600 | } else { | |
601 | iov_iter_advance(to, 0); | |
602 | } | |
603 | ceph_put_page_vector(pages, num_pages, false); | |
604 | } else { | |
605 | num_pages = calc_pages_for(off, len); | |
606 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); | |
607 | if (IS_ERR(pages)) | |
608 | return PTR_ERR(pages); | |
609 | ||
610 | ret = striped_read(inode, off, len, pages, num_pages, | |
611 | (off & ~PAGE_MASK), checkeof); | |
612 | if (ret > 0) { | |
613 | int l, k = 0; | |
614 | size_t left = ret; | |
615 | ||
616 | while (left) { | |
617 | size_t page_off = off & ~PAGE_MASK; | |
618 | size_t copy = min_t(size_t, left, | |
619 | PAGE_SIZE - page_off); | |
620 | l = copy_page_to_iter(pages[k++], page_off, | |
621 | copy, to); | |
622 | off += l; | |
623 | left -= l; | |
624 | if (l < copy) | |
625 | break; | |
626 | } | |
8eb4efb0 | 627 | } |
7ce469a5 | 628 | ceph_release_page_vector(pages, num_pages); |
8eb4efb0 | 629 | } |
124e68e7 | 630 | |
8eb4efb0 | 631 | if (off > iocb->ki_pos) { |
632 | ret = off - iocb->ki_pos; | |
633 | iocb->ki_pos = off; | |
634 | } | |
124e68e7 | 635 | |
7ce469a5 | 636 | dout("sync_read result %zd\n", ret); |
124e68e7 SW |
637 | return ret; |
638 | } | |
639 | ||
c8fe9b17 YZ |
640 | struct ceph_aio_request { |
641 | struct kiocb *iocb; | |
642 | size_t total_len; | |
5cabd8ac YZ |
643 | bool write; |
644 | bool should_dirty; | |
c8fe9b17 YZ |
645 | int error; |
646 | struct list_head osd_reqs; | |
647 | unsigned num_reqs; | |
648 | atomic_t pending_reqs; | |
5be0389d | 649 | struct timespec mtime; |
c8fe9b17 YZ |
650 | struct ceph_cap_flush *prealloc_cf; |
651 | }; | |
652 | ||
5be0389d YZ |
653 | struct ceph_aio_work { |
654 | struct work_struct work; | |
655 | struct ceph_osd_request *req; | |
656 | }; | |
657 | ||
658 | static void ceph_aio_retry_work(struct work_struct *work); | |
659 | ||
c8fe9b17 YZ |
660 | static void ceph_aio_complete(struct inode *inode, |
661 | struct ceph_aio_request *aio_req) | |
662 | { | |
663 | struct ceph_inode_info *ci = ceph_inode(inode); | |
664 | int ret; | |
665 | ||
666 | if (!atomic_dec_and_test(&aio_req->pending_reqs)) | |
667 | return; | |
668 | ||
669 | ret = aio_req->error; | |
670 | if (!ret) | |
671 | ret = aio_req->total_len; | |
672 | ||
673 | dout("ceph_aio_complete %p rc %d\n", inode, ret); | |
674 | ||
675 | if (ret >= 0 && aio_req->write) { | |
676 | int dirty; | |
677 | ||
678 | loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len; | |
679 | if (endoff > i_size_read(inode)) { | |
680 | if (ceph_inode_set_size(inode, endoff)) | |
681 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); | |
682 | } | |
683 | ||
684 | spin_lock(&ci->i_ceph_lock); | |
685 | ci->i_inline_version = CEPH_INLINE_NONE; | |
686 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, | |
687 | &aio_req->prealloc_cf); | |
688 | spin_unlock(&ci->i_ceph_lock); | |
689 | if (dirty) | |
690 | __mark_inode_dirty(inode, dirty); | |
691 | ||
692 | } | |
693 | ||
694 | ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR : | |
695 | CEPH_CAP_FILE_RD)); | |
696 | ||
697 | aio_req->iocb->ki_complete(aio_req->iocb, ret, 0); | |
698 | ||
699 | ceph_free_cap_flush(aio_req->prealloc_cf); | |
700 | kfree(aio_req); | |
701 | } | |
702 | ||
85e084fe | 703 | static void ceph_aio_complete_req(struct ceph_osd_request *req) |
c8fe9b17 YZ |
704 | { |
705 | int rc = req->r_result; | |
706 | struct inode *inode = req->r_inode; | |
707 | struct ceph_aio_request *aio_req = req->r_priv; | |
708 | struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0); | |
709 | int num_pages = calc_pages_for((u64)osd_data->alignment, | |
710 | osd_data->length); | |
711 | ||
712 | dout("ceph_aio_complete_req %p rc %d bytes %llu\n", | |
713 | inode, rc, osd_data->length); | |
714 | ||
715 | if (rc == -EOLDSNAPC) { | |
5be0389d YZ |
716 | struct ceph_aio_work *aio_work; |
717 | BUG_ON(!aio_req->write); | |
718 | ||
719 | aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS); | |
720 | if (aio_work) { | |
721 | INIT_WORK(&aio_work->work, ceph_aio_retry_work); | |
722 | aio_work->req = req; | |
723 | queue_work(ceph_inode_to_client(inode)->wb_wq, | |
724 | &aio_work->work); | |
725 | return; | |
726 | } | |
727 | rc = -ENOMEM; | |
728 | } else if (!aio_req->write) { | |
c8fe9b17 YZ |
729 | if (rc == -ENOENT) |
730 | rc = 0; | |
731 | if (rc >= 0 && osd_data->length > rc) { | |
732 | int zoff = osd_data->alignment + rc; | |
733 | int zlen = osd_data->length - rc; | |
734 | /* | |
735 | * If read is satisfied by single OSD request, | |
736 | * it can pass EOF. Otherwise read is within | |
737 | * i_size. | |
738 | */ | |
739 | if (aio_req->num_reqs == 1) { | |
740 | loff_t i_size = i_size_read(inode); | |
741 | loff_t endoff = aio_req->iocb->ki_pos + rc; | |
742 | if (endoff < i_size) | |
743 | zlen = min_t(size_t, zlen, | |
744 | i_size - endoff); | |
745 | aio_req->total_len = rc + zlen; | |
746 | } | |
747 | ||
748 | if (zlen > 0) | |
749 | ceph_zero_page_vector_range(zoff, zlen, | |
750 | osd_data->pages); | |
751 | } | |
752 | } | |
753 | ||
5cabd8ac | 754 | ceph_put_page_vector(osd_data->pages, num_pages, aio_req->should_dirty); |
c8fe9b17 YZ |
755 | ceph_osdc_put_request(req); |
756 | ||
757 | if (rc < 0) | |
758 | cmpxchg(&aio_req->error, 0, rc); | |
759 | ||
760 | ceph_aio_complete(inode, aio_req); | |
761 | return; | |
762 | } | |
763 | ||
5be0389d YZ |
764 | static void ceph_aio_retry_work(struct work_struct *work) |
765 | { | |
766 | struct ceph_aio_work *aio_work = | |
767 | container_of(work, struct ceph_aio_work, work); | |
768 | struct ceph_osd_request *orig_req = aio_work->req; | |
769 | struct ceph_aio_request *aio_req = orig_req->r_priv; | |
770 | struct inode *inode = orig_req->r_inode; | |
771 | struct ceph_inode_info *ci = ceph_inode(inode); | |
772 | struct ceph_snap_context *snapc; | |
773 | struct ceph_osd_request *req; | |
774 | int ret; | |
775 | ||
776 | spin_lock(&ci->i_ceph_lock); | |
777 | if (__ceph_have_pending_cap_snap(ci)) { | |
778 | struct ceph_cap_snap *capsnap = | |
779 | list_last_entry(&ci->i_cap_snaps, | |
780 | struct ceph_cap_snap, | |
781 | ci_item); | |
782 | snapc = ceph_get_snap_context(capsnap->context); | |
783 | } else { | |
784 | BUG_ON(!ci->i_head_snapc); | |
785 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
786 | } | |
787 | spin_unlock(&ci->i_ceph_lock); | |
788 | ||
789 | req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 2, | |
790 | false, GFP_NOFS); | |
1418bf07 DC |
791 | if (!req) { |
792 | ret = -ENOMEM; | |
5be0389d YZ |
793 | req = orig_req; |
794 | goto out; | |
795 | } | |
796 | ||
b178cf43 | 797 | req->r_flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE; |
63244fa1 | 798 | ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc); |
d30291b9 | 799 | ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid); |
5be0389d | 800 | |
13d1ad16 ID |
801 | ret = ceph_osdc_alloc_messages(req, GFP_NOFS); |
802 | if (ret) { | |
803 | ceph_osdc_put_request(req); | |
804 | req = orig_req; | |
805 | goto out; | |
806 | } | |
5be0389d YZ |
807 | |
808 | req->r_ops[0] = orig_req->r_ops[0]; | |
5be0389d | 809 | |
bb873b53 ID |
810 | req->r_mtime = aio_req->mtime; |
811 | req->r_data_offset = req->r_ops[0].extent.offset; | |
5be0389d | 812 | |
5be0389d YZ |
813 | ceph_osdc_put_request(orig_req); |
814 | ||
815 | req->r_callback = ceph_aio_complete_req; | |
816 | req->r_inode = inode; | |
817 | req->r_priv = aio_req; | |
a1f4020a | 818 | req->r_abort_on_full = true; |
5be0389d YZ |
819 | |
820 | ret = ceph_osdc_start_request(req->r_osdc, req, false); | |
821 | out: | |
822 | if (ret < 0) { | |
5be0389d | 823 | req->r_result = ret; |
85e084fe | 824 | ceph_aio_complete_req(req); |
5be0389d YZ |
825 | } |
826 | ||
db6aed70 | 827 | ceph_put_snap_context(snapc); |
5be0389d YZ |
828 | kfree(aio_work); |
829 | } | |
830 | ||
e8344e66 | 831 | static ssize_t |
c8fe9b17 YZ |
832 | ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter, |
833 | struct ceph_snap_context *snapc, | |
834 | struct ceph_cap_flush **pcf) | |
124e68e7 | 835 | { |
e8344e66 | 836 | struct file *file = iocb->ki_filp; |
496ad9aa | 837 | struct inode *inode = file_inode(file); |
124e68e7 | 838 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 | 839 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
acead002 | 840 | struct ceph_vino vino; |
124e68e7 SW |
841 | struct ceph_osd_request *req; |
842 | struct page **pages; | |
c8fe9b17 YZ |
843 | struct ceph_aio_request *aio_req = NULL; |
844 | int num_pages = 0; | |
124e68e7 | 845 | int flags; |
124e68e7 | 846 | int ret; |
c2050a45 | 847 | struct timespec mtime = current_time(inode); |
c8fe9b17 YZ |
848 | size_t count = iov_iter_count(iter); |
849 | loff_t pos = iocb->ki_pos; | |
850 | bool write = iov_iter_rw(iter) == WRITE; | |
5cabd8ac | 851 | bool should_dirty = !write && iter_is_iovec(iter); |
124e68e7 | 852 | |
c8fe9b17 | 853 | if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
124e68e7 SW |
854 | return -EROFS; |
855 | ||
1c0a9c2d YZ |
856 | dout("sync_direct_%s on file %p %lld~%u snapc %p seq %lld\n", |
857 | (write ? "write" : "read"), file, pos, (unsigned)count, | |
858 | snapc, snapc->seq); | |
124e68e7 | 859 | |
e8344e66 | 860 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); |
29065a51 YS |
861 | if (ret < 0) |
862 | return ret; | |
863 | ||
c8fe9b17 | 864 | if (write) { |
5d7eb1a3 | 865 | int ret2 = invalidate_inode_pages2_range(inode->i_mapping, |
09cbfeaf KS |
866 | pos >> PAGE_SHIFT, |
867 | (pos + count) >> PAGE_SHIFT); | |
5d7eb1a3 | 868 | if (ret2 < 0) |
a380a031 | 869 | dout("invalidate_inode_pages2_range returned %d\n", ret2); |
29065a51 | 870 | |
b178cf43 | 871 | flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE; |
c8fe9b17 YZ |
872 | } else { |
873 | flags = CEPH_OSD_FLAG_READ; | |
874 | } | |
124e68e7 | 875 | |
c8fe9b17 YZ |
876 | while (iov_iter_count(iter) > 0) { |
877 | u64 size = dio_get_pagev_size(iter); | |
878 | size_t start = 0; | |
879 | ssize_t len; | |
e8344e66 | 880 | |
d54fef90 ID |
881 | if (write) |
882 | size = min_t(u64, size, fsc->mount_options->wsize); | |
883 | else | |
884 | size = min_t(u64, size, fsc->mount_options->rsize); | |
885 | ||
e8344e66 | 886 | vino = ceph_vino(inode); |
887 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
c8fe9b17 | 888 | vino, pos, &size, 0, |
3fb99d48 | 889 | 1, |
c8fe9b17 YZ |
890 | write ? CEPH_OSD_OP_WRITE : |
891 | CEPH_OSD_OP_READ, | |
892 | flags, snapc, | |
e8344e66 | 893 | ci->i_truncate_seq, |
894 | ci->i_truncate_size, | |
895 | false); | |
896 | if (IS_ERR(req)) { | |
897 | ret = PTR_ERR(req); | |
eab87235 | 898 | break; |
e8344e66 | 899 | } |
124e68e7 | 900 | |
c8fe9b17 YZ |
901 | len = size; |
902 | pages = dio_get_pages_alloc(iter, len, &start, &num_pages); | |
b5b98989 | 903 | if (IS_ERR(pages)) { |
64c31311 | 904 | ceph_osdc_put_request(req); |
b5b98989 | 905 | ret = PTR_ERR(pages); |
64c31311 | 906 | break; |
124e68e7 SW |
907 | } |
908 | ||
909 | /* | |
c8fe9b17 YZ |
910 | * To simplify error handling, allow AIO when IO within i_size |
911 | * or IO can be satisfied by single OSD request. | |
124e68e7 | 912 | */ |
c8fe9b17 YZ |
913 | if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) && |
914 | (len == count || pos + count <= i_size_read(inode))) { | |
915 | aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL); | |
916 | if (aio_req) { | |
917 | aio_req->iocb = iocb; | |
918 | aio_req->write = write; | |
5cabd8ac | 919 | aio_req->should_dirty = should_dirty; |
c8fe9b17 YZ |
920 | INIT_LIST_HEAD(&aio_req->osd_reqs); |
921 | if (write) { | |
5be0389d | 922 | aio_req->mtime = mtime; |
c8fe9b17 YZ |
923 | swap(aio_req->prealloc_cf, *pcf); |
924 | } | |
925 | } | |
926 | /* ignore error */ | |
927 | } | |
928 | ||
929 | if (write) { | |
930 | /* | |
931 | * throw out any page cache pages in this range. this | |
932 | * may block. | |
933 | */ | |
934 | truncate_inode_pages_range(inode->i_mapping, pos, | |
09cbfeaf | 935 | (pos+len) | (PAGE_SIZE - 1)); |
c8fe9b17 | 936 | |
bb873b53 | 937 | req->r_mtime = mtime; |
c8fe9b17 YZ |
938 | } |
939 | ||
c8fe9b17 YZ |
940 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, start, |
941 | false, false); | |
e8344e66 | 942 | |
c8fe9b17 YZ |
943 | if (aio_req) { |
944 | aio_req->total_len += len; | |
945 | aio_req->num_reqs++; | |
946 | atomic_inc(&aio_req->pending_reqs); | |
947 | ||
948 | req->r_callback = ceph_aio_complete_req; | |
949 | req->r_inode = inode; | |
950 | req->r_priv = aio_req; | |
951 | list_add_tail(&req->r_unsafe_item, &aio_req->osd_reqs); | |
952 | ||
953 | pos += len; | |
954 | iov_iter_advance(iter, len); | |
955 | continue; | |
956 | } | |
957 | ||
958 | ret = ceph_osdc_start_request(req->r_osdc, req, false); | |
e8344e66 | 959 | if (!ret) |
960 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
961 | ||
c8fe9b17 YZ |
962 | size = i_size_read(inode); |
963 | if (!write) { | |
964 | if (ret == -ENOENT) | |
965 | ret = 0; | |
966 | if (ret >= 0 && ret < len && pos + ret < size) { | |
967 | int zlen = min_t(size_t, len - ret, | |
968 | size - pos - ret); | |
969 | ceph_zero_page_vector_range(start + ret, zlen, | |
970 | pages); | |
971 | ret += zlen; | |
972 | } | |
973 | if (ret >= 0) | |
974 | len = ret; | |
975 | } | |
976 | ||
5cabd8ac | 977 | ceph_put_page_vector(pages, num_pages, should_dirty); |
e8344e66 | 978 | |
e8344e66 | 979 | ceph_osdc_put_request(req); |
c8fe9b17 | 980 | if (ret < 0) |
e8344e66 | 981 | break; |
64c31311 | 982 | |
c8fe9b17 YZ |
983 | pos += len; |
984 | iov_iter_advance(iter, len); | |
985 | ||
986 | if (!write && pos >= size) | |
e8344e66 | 987 | break; |
64c31311 | 988 | |
c8fe9b17 YZ |
989 | if (write && pos > size) { |
990 | if (ceph_inode_set_size(inode, pos)) | |
64c31311 AV |
991 | ceph_check_caps(ceph_inode(inode), |
992 | CHECK_CAPS_AUTHONLY, | |
993 | NULL); | |
994 | } | |
e8344e66 | 995 | } |
996 | ||
c8fe9b17 | 997 | if (aio_req) { |
fc8c3892 YZ |
998 | LIST_HEAD(osd_reqs); |
999 | ||
c8fe9b17 YZ |
1000 | if (aio_req->num_reqs == 0) { |
1001 | kfree(aio_req); | |
1002 | return ret; | |
1003 | } | |
1004 | ||
1005 | ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR : | |
1006 | CEPH_CAP_FILE_RD); | |
1007 | ||
fc8c3892 YZ |
1008 | list_splice(&aio_req->osd_reqs, &osd_reqs); |
1009 | while (!list_empty(&osd_reqs)) { | |
1010 | req = list_first_entry(&osd_reqs, | |
c8fe9b17 YZ |
1011 | struct ceph_osd_request, |
1012 | r_unsafe_item); | |
1013 | list_del_init(&req->r_unsafe_item); | |
1014 | if (ret >= 0) | |
1015 | ret = ceph_osdc_start_request(req->r_osdc, | |
1016 | req, false); | |
1017 | if (ret < 0) { | |
1018 | req->r_result = ret; | |
85e084fe | 1019 | ceph_aio_complete_req(req); |
c8fe9b17 YZ |
1020 | } |
1021 | } | |
1022 | return -EIOCBQUEUED; | |
1023 | } | |
1024 | ||
1025 | if (ret != -EOLDSNAPC && pos > iocb->ki_pos) { | |
1026 | ret = pos - iocb->ki_pos; | |
e8344e66 | 1027 | iocb->ki_pos = pos; |
e8344e66 | 1028 | } |
1029 | return ret; | |
1030 | } | |
1031 | ||
e8344e66 | 1032 | /* |
1033 | * Synchronous write, straight from __user pointer or user pages. | |
1034 | * | |
1035 | * If write spans object boundary, just do multiple writes. (For a | |
1036 | * correct atomic write, we should e.g. take write locks on all | |
1037 | * objects, rollback on failure, etc.) | |
1038 | */ | |
06fee30f | 1039 | static ssize_t |
5dda377c YZ |
1040 | ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos, |
1041 | struct ceph_snap_context *snapc) | |
e8344e66 | 1042 | { |
1043 | struct file *file = iocb->ki_filp; | |
1044 | struct inode *inode = file_inode(file); | |
1045 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1046 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
e8344e66 | 1047 | struct ceph_vino vino; |
1048 | struct ceph_osd_request *req; | |
1049 | struct page **pages; | |
1050 | u64 len; | |
1051 | int num_pages; | |
1052 | int written = 0; | |
1053 | int flags; | |
e8344e66 | 1054 | int ret; |
efb0ca76 | 1055 | bool check_caps = false; |
c2050a45 | 1056 | struct timespec mtime = current_time(inode); |
4908b822 | 1057 | size_t count = iov_iter_count(from); |
e8344e66 | 1058 | |
1059 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) | |
1060 | return -EROFS; | |
1061 | ||
1c0a9c2d YZ |
1062 | dout("sync_write on file %p %lld~%u snapc %p seq %lld\n", |
1063 | file, pos, (unsigned)count, snapc, snapc->seq); | |
e8344e66 | 1064 | |
1065 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); | |
1066 | if (ret < 0) | |
1067 | return ret; | |
1068 | ||
1069 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
09cbfeaf KS |
1070 | pos >> PAGE_SHIFT, |
1071 | (pos + count) >> PAGE_SHIFT); | |
e8344e66 | 1072 | if (ret < 0) |
1073 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
1074 | ||
b178cf43 | 1075 | flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE; |
e8344e66 | 1076 | |
4908b822 | 1077 | while ((len = iov_iter_count(from)) > 0) { |
e8344e66 | 1078 | size_t left; |
1079 | int n; | |
1080 | ||
e8344e66 | 1081 | vino = ceph_vino(inode); |
1082 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
715e4cd4 | 1083 | vino, pos, &len, 0, 1, |
e8344e66 | 1084 | CEPH_OSD_OP_WRITE, flags, snapc, |
1085 | ci->i_truncate_seq, | |
1086 | ci->i_truncate_size, | |
1087 | false); | |
1088 | if (IS_ERR(req)) { | |
1089 | ret = PTR_ERR(req); | |
eab87235 | 1090 | break; |
e8344e66 | 1091 | } |
1092 | ||
1093 | /* | |
1094 | * write from beginning of first page, | |
1095 | * regardless of io alignment | |
1096 | */ | |
09cbfeaf | 1097 | num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
e8344e66 | 1098 | |
687265e5 | 1099 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
124e68e7 SW |
1100 | if (IS_ERR(pages)) { |
1101 | ret = PTR_ERR(pages); | |
1102 | goto out; | |
1103 | } | |
e8344e66 | 1104 | |
1105 | left = len; | |
1106 | for (n = 0; n < num_pages; n++) { | |
125d725c | 1107 | size_t plen = min_t(size_t, left, PAGE_SIZE); |
4908b822 | 1108 | ret = copy_page_from_iter(pages[n], 0, plen, from); |
e8344e66 | 1109 | if (ret != plen) { |
1110 | ret = -EFAULT; | |
1111 | break; | |
1112 | } | |
1113 | left -= ret; | |
e8344e66 | 1114 | } |
1115 | ||
124e68e7 SW |
1116 | if (ret < 0) { |
1117 | ceph_release_page_vector(pages, num_pages); | |
1118 | goto out; | |
1119 | } | |
1120 | ||
e8344e66 | 1121 | req->r_inode = inode; |
124e68e7 | 1122 | |
e8344e66 | 1123 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, |
1124 | false, true); | |
02ee07d3 | 1125 | |
bb873b53 | 1126 | req->r_mtime = mtime; |
e8344e66 | 1127 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
1128 | if (!ret) | |
1129 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
124e68e7 SW |
1130 | |
1131 | out: | |
e8344e66 | 1132 | ceph_osdc_put_request(req); |
26544c62 JL |
1133 | if (ret != 0) { |
1134 | ceph_set_error_write(ci); | |
e8344e66 | 1135 | break; |
26544c62 JL |
1136 | } |
1137 | ||
1138 | ceph_clear_error_write(ci); | |
1139 | pos += len; | |
1140 | written += len; | |
1141 | if (pos > i_size_read(inode)) { | |
1142 | check_caps = ceph_inode_set_size(inode, pos); | |
1143 | if (check_caps) | |
1144 | ceph_check_caps(ceph_inode(inode), | |
1145 | CHECK_CAPS_AUTHONLY, | |
1146 | NULL); | |
1147 | } | |
1148 | ||
e8344e66 | 1149 | } |
124e68e7 | 1150 | |
e8344e66 | 1151 | if (ret != -EOLDSNAPC && written > 0) { |
124e68e7 | 1152 | ret = written; |
e8344e66 | 1153 | iocb->ki_pos = pos; |
124e68e7 SW |
1154 | } |
1155 | return ret; | |
1156 | } | |
1157 | ||
1158 | /* | |
1159 | * Wrap generic_file_aio_read with checks for cap bits on the inode. | |
1160 | * Atomically grab references, so that those bits are not released | |
1161 | * back to the MDS mid-read. | |
1162 | * | |
1163 | * Hmm, the sync read case isn't actually async... should it be? | |
1164 | */ | |
3644424d | 1165 | static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) |
124e68e7 SW |
1166 | { |
1167 | struct file *filp = iocb->ki_filp; | |
2962507c | 1168 | struct ceph_file_info *fi = filp->private_data; |
66ee59af | 1169 | size_t len = iov_iter_count(to); |
496ad9aa | 1170 | struct inode *inode = file_inode(filp); |
124e68e7 | 1171 | struct ceph_inode_info *ci = ceph_inode(inode); |
3738daa6 | 1172 | struct page *pinned_page = NULL; |
124e68e7 | 1173 | ssize_t ret; |
2962507c | 1174 | int want, got = 0; |
83701246 | 1175 | int retry_op = 0, read = 0; |
124e68e7 | 1176 | |
6a026589 | 1177 | again: |
8eb4efb0 | 1178 | dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n", |
1179 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode); | |
1180 | ||
2962507c SW |
1181 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
1182 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
1183 | else | |
1184 | want = CEPH_CAP_FILE_CACHE; | |
3738daa6 | 1185 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page); |
124e68e7 | 1186 | if (ret < 0) |
8eb4efb0 | 1187 | return ret; |
124e68e7 | 1188 | |
2962507c | 1189 | if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 || |
2ba48ce5 | 1190 | (iocb->ki_flags & IOCB_DIRECT) || |
8eb4efb0 | 1191 | (fi->flags & CEPH_F_SYNC)) { |
8eb4efb0 | 1192 | |
1193 | dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n", | |
1194 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, | |
1195 | ceph_cap_string(got)); | |
1196 | ||
83701246 | 1197 | if (ci->i_inline_version == CEPH_INLINE_NONE) { |
c8fe9b17 YZ |
1198 | if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) { |
1199 | ret = ceph_direct_read_write(iocb, to, | |
1200 | NULL, NULL); | |
1201 | if (ret >= 0 && ret < len) | |
1202 | retry_op = CHECK_EOF; | |
1203 | } else { | |
1204 | ret = ceph_sync_read(iocb, to, &retry_op); | |
1205 | } | |
83701246 YZ |
1206 | } else { |
1207 | retry_op = READ_INLINE; | |
1208 | } | |
8eb4efb0 | 1209 | } else { |
3bd4d543 | 1210 | CEPH_DEFINE_RW_CONTEXT(rw_ctx, got); |
8eb4efb0 | 1211 | dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n", |
3644424d | 1212 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, |
8eb4efb0 | 1213 | ceph_cap_string(got)); |
3bd4d543 | 1214 | ceph_add_rw_context(fi, &rw_ctx); |
3644424d | 1215 | ret = generic_file_read_iter(iocb, to); |
3bd4d543 | 1216 | ceph_del_rw_context(fi, &rw_ctx); |
8eb4efb0 | 1217 | } |
124e68e7 SW |
1218 | dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", |
1219 | inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); | |
3738daa6 | 1220 | if (pinned_page) { |
09cbfeaf | 1221 | put_page(pinned_page); |
3738daa6 YZ |
1222 | pinned_page = NULL; |
1223 | } | |
124e68e7 | 1224 | ceph_put_cap_refs(ci, got); |
c8fe9b17 | 1225 | if (retry_op > HAVE_RETRIED && ret >= 0) { |
83701246 YZ |
1226 | int statret; |
1227 | struct page *page = NULL; | |
1228 | loff_t i_size; | |
1229 | if (retry_op == READ_INLINE) { | |
687265e5 | 1230 | page = __page_cache_alloc(GFP_KERNEL); |
83701246 YZ |
1231 | if (!page) |
1232 | return -ENOMEM; | |
1233 | } | |
6a026589 | 1234 | |
83701246 YZ |
1235 | statret = __ceph_do_getattr(inode, page, |
1236 | CEPH_STAT_CAP_INLINE_DATA, !!page); | |
1237 | if (statret < 0) { | |
0d7718f6 NB |
1238 | if (page) |
1239 | __free_page(page); | |
83701246 YZ |
1240 | if (statret == -ENODATA) { |
1241 | BUG_ON(retry_op != READ_INLINE); | |
1242 | goto again; | |
1243 | } | |
1244 | return statret; | |
1245 | } | |
6a026589 | 1246 | |
83701246 YZ |
1247 | i_size = i_size_read(inode); |
1248 | if (retry_op == READ_INLINE) { | |
fcc02d2a YZ |
1249 | BUG_ON(ret > 0 || read > 0); |
1250 | if (iocb->ki_pos < i_size && | |
09cbfeaf | 1251 | iocb->ki_pos < PAGE_SIZE) { |
83701246 YZ |
1252 | loff_t end = min_t(loff_t, i_size, |
1253 | iocb->ki_pos + len); | |
09cbfeaf | 1254 | end = min_t(loff_t, end, PAGE_SIZE); |
83701246 YZ |
1255 | if (statret < end) |
1256 | zero_user_segment(page, statret, end); | |
1257 | ret = copy_page_to_iter(page, | |
1258 | iocb->ki_pos & ~PAGE_MASK, | |
1259 | end - iocb->ki_pos, to); | |
1260 | iocb->ki_pos += ret; | |
fcc02d2a YZ |
1261 | read += ret; |
1262 | } | |
1263 | if (iocb->ki_pos < i_size && read < len) { | |
1264 | size_t zlen = min_t(size_t, len - read, | |
1265 | i_size - iocb->ki_pos); | |
1266 | ret = iov_iter_zero(zlen, to); | |
1267 | iocb->ki_pos += ret; | |
1268 | read += ret; | |
83701246 YZ |
1269 | } |
1270 | __free_pages(page, 0); | |
fcc02d2a | 1271 | return read; |
83701246 | 1272 | } |
6a026589 SW |
1273 | |
1274 | /* hit EOF or hole? */ | |
83701246 | 1275 | if (retry_op == CHECK_EOF && iocb->ki_pos < i_size && |
fcc02d2a | 1276 | ret < len) { |
8eb4efb0 | 1277 | dout("sync_read hit hole, ppos %lld < size %lld" |
99c88e69 | 1278 | ", reading more\n", iocb->ki_pos, i_size); |
8eb4efb0 | 1279 | |
6a026589 | 1280 | read += ret; |
6a026589 | 1281 | len -= ret; |
c8fe9b17 | 1282 | retry_op = HAVE_RETRIED; |
6a026589 SW |
1283 | goto again; |
1284 | } | |
1285 | } | |
8eb4efb0 | 1286 | |
6a026589 SW |
1287 | if (ret >= 0) |
1288 | ret += read; | |
1289 | ||
124e68e7 SW |
1290 | return ret; |
1291 | } | |
1292 | ||
1293 | /* | |
1294 | * Take cap references to avoid releasing caps to MDS mid-write. | |
1295 | * | |
1296 | * If we are synchronous, and write with an old snap context, the OSD | |
1297 | * may return EOLDSNAPC. In that case, retry the write.. _after_ | |
1298 | * dropping our cap refs and allowing the pending snap to logically | |
1299 | * complete _before_ this write occurs. | |
1300 | * | |
1301 | * If we are near ENOSPC, write synchronously. | |
1302 | */ | |
4908b822 | 1303 | static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from) |
124e68e7 SW |
1304 | { |
1305 | struct file *file = iocb->ki_filp; | |
33caad32 | 1306 | struct ceph_file_info *fi = file->private_data; |
496ad9aa | 1307 | struct inode *inode = file_inode(file); |
124e68e7 | 1308 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 YS |
1309 | struct ceph_osd_client *osdc = |
1310 | &ceph_sb_to_client(inode->i_sb)->client->osdc; | |
f66fd9f0 | 1311 | struct ceph_cap_flush *prealloc_cf; |
3309dd04 | 1312 | ssize_t count, written = 0; |
03d254ed | 1313 | int err, want, got; |
3309dd04 | 1314 | loff_t pos; |
124e68e7 SW |
1315 | |
1316 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
1317 | return -EROFS; | |
1318 | ||
f66fd9f0 YZ |
1319 | prealloc_cf = ceph_alloc_cap_flush(); |
1320 | if (!prealloc_cf) | |
1321 | return -ENOMEM; | |
1322 | ||
a5cd74ad | 1323 | retry_snap: |
5955102c | 1324 | inode_lock(inode); |
03d254ed | 1325 | |
03d254ed | 1326 | /* We can write back this queue in page reclaim */ |
de1414a6 | 1327 | current->backing_dev_info = inode_to_bdi(inode); |
03d254ed | 1328 | |
55b0b31c YZ |
1329 | if (iocb->ki_flags & IOCB_APPEND) { |
1330 | err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); | |
1331 | if (err < 0) | |
1332 | goto out; | |
1333 | } | |
1334 | ||
3309dd04 AV |
1335 | err = generic_write_checks(iocb, from); |
1336 | if (err <= 0) | |
03d254ed YZ |
1337 | goto out; |
1338 | ||
3309dd04 AV |
1339 | pos = iocb->ki_pos; |
1340 | count = iov_iter_count(from); | |
5fa8e0a1 | 1341 | err = file_remove_privs(file); |
03d254ed YZ |
1342 | if (err) |
1343 | goto out; | |
1344 | ||
1345 | err = file_update_time(file); | |
1346 | if (err) | |
1347 | goto out; | |
1348 | ||
28127bdd YZ |
1349 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
1350 | err = ceph_uninline_data(file, NULL); | |
1351 | if (err < 0) | |
1352 | goto out; | |
1353 | } | |
1354 | ||
26544c62 | 1355 | /* FIXME: not complete since it doesn't account for being at quota */ |
b7ec35b3 | 1356 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL)) { |
03d254ed | 1357 | err = -ENOSPC; |
6070e0c1 YZ |
1358 | goto out; |
1359 | } | |
03d254ed | 1360 | |
ac7f29bf | 1361 | dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n", |
99c88e69 | 1362 | inode, ceph_vinop(inode), pos, count, i_size_read(inode)); |
7971bd92 SW |
1363 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
1364 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1365 | else | |
1366 | want = CEPH_CAP_FILE_BUFFER; | |
03d254ed | 1367 | got = 0; |
3738daa6 YZ |
1368 | err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count, |
1369 | &got, NULL); | |
03d254ed | 1370 | if (err < 0) |
37505d57 | 1371 | goto out; |
124e68e7 | 1372 | |
ac7f29bf | 1373 | dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n", |
03d254ed | 1374 | inode, ceph_vinop(inode), pos, count, ceph_cap_string(got)); |
7971bd92 SW |
1375 | |
1376 | if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || | |
26544c62 JL |
1377 | (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC) || |
1378 | (ci->i_ceph_flags & CEPH_I_ERROR_WRITE)) { | |
5dda377c | 1379 | struct ceph_snap_context *snapc; |
4908b822 | 1380 | struct iov_iter data; |
5955102c | 1381 | inode_unlock(inode); |
5dda377c YZ |
1382 | |
1383 | spin_lock(&ci->i_ceph_lock); | |
1384 | if (__ceph_have_pending_cap_snap(ci)) { | |
1385 | struct ceph_cap_snap *capsnap = | |
1386 | list_last_entry(&ci->i_cap_snaps, | |
1387 | struct ceph_cap_snap, | |
1388 | ci_item); | |
1389 | snapc = ceph_get_snap_context(capsnap->context); | |
1390 | } else { | |
1391 | BUG_ON(!ci->i_head_snapc); | |
1392 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
1393 | } | |
1394 | spin_unlock(&ci->i_ceph_lock); | |
1395 | ||
4908b822 AV |
1396 | /* we might need to revert back to that point */ |
1397 | data = *from; | |
2ba48ce5 | 1398 | if (iocb->ki_flags & IOCB_DIRECT) |
c8fe9b17 YZ |
1399 | written = ceph_direct_read_write(iocb, &data, snapc, |
1400 | &prealloc_cf); | |
e8344e66 | 1401 | else |
5dda377c | 1402 | written = ceph_sync_write(iocb, &data, pos, snapc); |
4908b822 AV |
1403 | if (written > 0) |
1404 | iov_iter_advance(from, written); | |
5dda377c | 1405 | ceph_put_snap_context(snapc); |
7971bd92 | 1406 | } else { |
b0d7c223 YZ |
1407 | /* |
1408 | * No need to acquire the i_truncate_mutex. Because | |
1409 | * the MDS revokes Fwb caps before sending truncate | |
1410 | * message to us. We can't get Fwb cap while there | |
1411 | * are pending vmtruncate. So write and vmtruncate | |
1412 | * can not run at the same time | |
1413 | */ | |
4908b822 | 1414 | written = generic_perform_write(file, from, pos); |
aec605f4 AV |
1415 | if (likely(written >= 0)) |
1416 | iocb->ki_pos = pos + written; | |
5955102c | 1417 | inode_unlock(inode); |
7971bd92 | 1418 | } |
d8de9ab6 | 1419 | |
03d254ed | 1420 | if (written >= 0) { |
fca65b4a | 1421 | int dirty; |
be655596 | 1422 | spin_lock(&ci->i_ceph_lock); |
28127bdd | 1423 | ci->i_inline_version = CEPH_INLINE_NONE; |
f66fd9f0 YZ |
1424 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
1425 | &prealloc_cf); | |
be655596 | 1426 | spin_unlock(&ci->i_ceph_lock); |
fca65b4a SW |
1427 | if (dirty) |
1428 | __mark_inode_dirty(inode, dirty); | |
124e68e7 | 1429 | } |
7971bd92 | 1430 | |
124e68e7 | 1431 | dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n", |
4908b822 | 1432 | inode, ceph_vinop(inode), pos, (unsigned)count, |
7971bd92 | 1433 | ceph_cap_string(got)); |
124e68e7 | 1434 | ceph_put_cap_refs(ci, got); |
7971bd92 | 1435 | |
a5cd74ad YZ |
1436 | if (written == -EOLDSNAPC) { |
1437 | dout("aio_write %p %llx.%llx %llu~%u" "got EOLDSNAPC, retrying\n", | |
1438 | inode, ceph_vinop(inode), pos, (unsigned)count); | |
1439 | goto retry_snap; | |
1440 | } | |
1441 | ||
6aa657c8 | 1442 | if (written >= 0) { |
b7ec35b3 | 1443 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_NEARFULL)) |
6aa657c8 | 1444 | iocb->ki_flags |= IOCB_DSYNC; |
6aa657c8 | 1445 | written = generic_write_sync(iocb, written); |
6070e0c1 | 1446 | } |
03d254ed | 1447 | |
2f75e9e1 SW |
1448 | goto out_unlocked; |
1449 | ||
03d254ed | 1450 | out: |
5955102c | 1451 | inode_unlock(inode); |
2f75e9e1 | 1452 | out_unlocked: |
f66fd9f0 | 1453 | ceph_free_cap_flush(prealloc_cf); |
03d254ed | 1454 | current->backing_dev_info = NULL; |
03d254ed | 1455 | return written ? written : err; |
124e68e7 SW |
1456 | } |
1457 | ||
1458 | /* | |
1459 | * llseek. be sure to verify file size on SEEK_END. | |
1460 | */ | |
965c8e59 | 1461 | static loff_t ceph_llseek(struct file *file, loff_t offset, int whence) |
124e68e7 SW |
1462 | { |
1463 | struct inode *inode = file->f_mapping->host; | |
99c88e69 | 1464 | loff_t i_size; |
955818cd | 1465 | loff_t ret; |
124e68e7 | 1466 | |
5955102c | 1467 | inode_lock(inode); |
6a82c47a | 1468 | |
965c8e59 | 1469 | if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { |
508b32d8 | 1470 | ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); |
955818cd | 1471 | if (ret < 0) |
124e68e7 | 1472 | goto out; |
06222e49 JB |
1473 | } |
1474 | ||
99c88e69 | 1475 | i_size = i_size_read(inode); |
965c8e59 | 1476 | switch (whence) { |
06222e49 | 1477 | case SEEK_END: |
99c88e69 | 1478 | offset += i_size; |
124e68e7 SW |
1479 | break; |
1480 | case SEEK_CUR: | |
1481 | /* | |
1482 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
1483 | * position-querying operation. Avoid rewriting the "same" | |
1484 | * f_pos value back to the file because a concurrent read(), | |
1485 | * write() or lseek() might have altered it | |
1486 | */ | |
1487 | if (offset == 0) { | |
955818cd | 1488 | ret = file->f_pos; |
124e68e7 SW |
1489 | goto out; |
1490 | } | |
1491 | offset += file->f_pos; | |
1492 | break; | |
06222e49 | 1493 | case SEEK_DATA: |
397f2389 | 1494 | if (offset < 0 || offset >= i_size) { |
06222e49 JB |
1495 | ret = -ENXIO; |
1496 | goto out; | |
1497 | } | |
1498 | break; | |
1499 | case SEEK_HOLE: | |
397f2389 | 1500 | if (offset < 0 || offset >= i_size) { |
06222e49 JB |
1501 | ret = -ENXIO; |
1502 | goto out; | |
1503 | } | |
99c88e69 | 1504 | offset = i_size; |
06222e49 | 1505 | break; |
124e68e7 SW |
1506 | } |
1507 | ||
955818cd | 1508 | ret = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
124e68e7 SW |
1509 | |
1510 | out: | |
5955102c | 1511 | inode_unlock(inode); |
955818cd | 1512 | return ret; |
124e68e7 SW |
1513 | } |
1514 | ||
ad7a60de LW |
1515 | static inline void ceph_zero_partial_page( |
1516 | struct inode *inode, loff_t offset, unsigned size) | |
1517 | { | |
1518 | struct page *page; | |
09cbfeaf | 1519 | pgoff_t index = offset >> PAGE_SHIFT; |
ad7a60de LW |
1520 | |
1521 | page = find_lock_page(inode->i_mapping, index); | |
1522 | if (page) { | |
1523 | wait_on_page_writeback(page); | |
09cbfeaf | 1524 | zero_user(page, offset & (PAGE_SIZE - 1), size); |
ad7a60de | 1525 | unlock_page(page); |
09cbfeaf | 1526 | put_page(page); |
ad7a60de LW |
1527 | } |
1528 | } | |
1529 | ||
1530 | static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset, | |
1531 | loff_t length) | |
1532 | { | |
09cbfeaf | 1533 | loff_t nearly = round_up(offset, PAGE_SIZE); |
ad7a60de LW |
1534 | if (offset < nearly) { |
1535 | loff_t size = nearly - offset; | |
1536 | if (length < size) | |
1537 | size = length; | |
1538 | ceph_zero_partial_page(inode, offset, size); | |
1539 | offset += size; | |
1540 | length -= size; | |
1541 | } | |
09cbfeaf KS |
1542 | if (length >= PAGE_SIZE) { |
1543 | loff_t size = round_down(length, PAGE_SIZE); | |
ad7a60de LW |
1544 | truncate_pagecache_range(inode, offset, offset + size - 1); |
1545 | offset += size; | |
1546 | length -= size; | |
1547 | } | |
1548 | if (length) | |
1549 | ceph_zero_partial_page(inode, offset, length); | |
1550 | } | |
1551 | ||
1552 | static int ceph_zero_partial_object(struct inode *inode, | |
1553 | loff_t offset, loff_t *length) | |
1554 | { | |
1555 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1556 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
1557 | struct ceph_osd_request *req; | |
1558 | int ret = 0; | |
1559 | loff_t zero = 0; | |
1560 | int op; | |
1561 | ||
1562 | if (!length) { | |
1563 | op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE; | |
1564 | length = &zero; | |
1565 | } else { | |
1566 | op = CEPH_OSD_OP_ZERO; | |
1567 | } | |
1568 | ||
1569 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1570 | ceph_vino(inode), | |
1571 | offset, length, | |
715e4cd4 | 1572 | 0, 1, op, |
54ea0046 | 1573 | CEPH_OSD_FLAG_WRITE, |
ad7a60de LW |
1574 | NULL, 0, 0, false); |
1575 | if (IS_ERR(req)) { | |
1576 | ret = PTR_ERR(req); | |
1577 | goto out; | |
1578 | } | |
1579 | ||
bb873b53 | 1580 | req->r_mtime = inode->i_mtime; |
ad7a60de LW |
1581 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
1582 | if (!ret) { | |
1583 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1584 | if (ret == -ENOENT) | |
1585 | ret = 0; | |
1586 | } | |
1587 | ceph_osdc_put_request(req); | |
1588 | ||
1589 | out: | |
1590 | return ret; | |
1591 | } | |
1592 | ||
1593 | static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length) | |
1594 | { | |
1595 | int ret = 0; | |
1596 | struct ceph_inode_info *ci = ceph_inode(inode); | |
7627151e YZ |
1597 | s32 stripe_unit = ci->i_layout.stripe_unit; |
1598 | s32 stripe_count = ci->i_layout.stripe_count; | |
1599 | s32 object_size = ci->i_layout.object_size; | |
b314a90d SW |
1600 | u64 object_set_size = object_size * stripe_count; |
1601 | u64 nearly, t; | |
1602 | ||
1603 | /* round offset up to next period boundary */ | |
1604 | nearly = offset + object_set_size - 1; | |
1605 | t = nearly; | |
1606 | nearly -= do_div(t, object_set_size); | |
ad7a60de | 1607 | |
ad7a60de LW |
1608 | while (length && offset < nearly) { |
1609 | loff_t size = length; | |
1610 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1611 | if (ret < 0) | |
1612 | return ret; | |
1613 | offset += size; | |
1614 | length -= size; | |
1615 | } | |
1616 | while (length >= object_set_size) { | |
1617 | int i; | |
1618 | loff_t pos = offset; | |
1619 | for (i = 0; i < stripe_count; ++i) { | |
1620 | ret = ceph_zero_partial_object(inode, pos, NULL); | |
1621 | if (ret < 0) | |
1622 | return ret; | |
1623 | pos += stripe_unit; | |
1624 | } | |
1625 | offset += object_set_size; | |
1626 | length -= object_set_size; | |
1627 | } | |
1628 | while (length) { | |
1629 | loff_t size = length; | |
1630 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1631 | if (ret < 0) | |
1632 | return ret; | |
1633 | offset += size; | |
1634 | length -= size; | |
1635 | } | |
1636 | return ret; | |
1637 | } | |
1638 | ||
1639 | static long ceph_fallocate(struct file *file, int mode, | |
1640 | loff_t offset, loff_t length) | |
1641 | { | |
1642 | struct ceph_file_info *fi = file->private_data; | |
aa8b60e0 | 1643 | struct inode *inode = file_inode(file); |
ad7a60de LW |
1644 | struct ceph_inode_info *ci = ceph_inode(inode); |
1645 | struct ceph_osd_client *osdc = | |
1646 | &ceph_inode_to_client(inode)->client->osdc; | |
f66fd9f0 | 1647 | struct ceph_cap_flush *prealloc_cf; |
ad7a60de LW |
1648 | int want, got = 0; |
1649 | int dirty; | |
1650 | int ret = 0; | |
1651 | loff_t endoff = 0; | |
1652 | loff_t size; | |
1653 | ||
494d77bf YZ |
1654 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
1655 | return -EOPNOTSUPP; | |
1656 | ||
ad7a60de LW |
1657 | if (!S_ISREG(inode->i_mode)) |
1658 | return -EOPNOTSUPP; | |
1659 | ||
f66fd9f0 YZ |
1660 | prealloc_cf = ceph_alloc_cap_flush(); |
1661 | if (!prealloc_cf) | |
1662 | return -ENOMEM; | |
1663 | ||
5955102c | 1664 | inode_lock(inode); |
ad7a60de LW |
1665 | |
1666 | if (ceph_snap(inode) != CEPH_NOSNAP) { | |
1667 | ret = -EROFS; | |
1668 | goto unlock; | |
1669 | } | |
1670 | ||
b7ec35b3 ID |
1671 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) && |
1672 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
ad7a60de LW |
1673 | ret = -ENOSPC; |
1674 | goto unlock; | |
1675 | } | |
1676 | ||
28127bdd YZ |
1677 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
1678 | ret = ceph_uninline_data(file, NULL); | |
1679 | if (ret < 0) | |
1680 | goto unlock; | |
1681 | } | |
1682 | ||
ad7a60de | 1683 | size = i_size_read(inode); |
42c99fc4 | 1684 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
ad7a60de | 1685 | endoff = offset + length; |
42c99fc4 LH |
1686 | ret = inode_newsize_ok(inode, endoff); |
1687 | if (ret) | |
1688 | goto unlock; | |
1689 | } | |
ad7a60de LW |
1690 | |
1691 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1692 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1693 | else | |
1694 | want = CEPH_CAP_FILE_BUFFER; | |
1695 | ||
3738daa6 | 1696 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL); |
ad7a60de LW |
1697 | if (ret < 0) |
1698 | goto unlock; | |
1699 | ||
1700 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1701 | if (offset < size) | |
1702 | ceph_zero_pagecache_range(inode, offset, length); | |
1703 | ret = ceph_zero_objects(inode, offset, length); | |
1704 | } else if (endoff > size) { | |
1705 | truncate_pagecache_range(inode, size, -1); | |
1706 | if (ceph_inode_set_size(inode, endoff)) | |
1707 | ceph_check_caps(ceph_inode(inode), | |
1708 | CHECK_CAPS_AUTHONLY, NULL); | |
1709 | } | |
1710 | ||
1711 | if (!ret) { | |
1712 | spin_lock(&ci->i_ceph_lock); | |
28127bdd | 1713 | ci->i_inline_version = CEPH_INLINE_NONE; |
f66fd9f0 YZ |
1714 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
1715 | &prealloc_cf); | |
ad7a60de LW |
1716 | spin_unlock(&ci->i_ceph_lock); |
1717 | if (dirty) | |
1718 | __mark_inode_dirty(inode, dirty); | |
1719 | } | |
1720 | ||
1721 | ceph_put_cap_refs(ci, got); | |
1722 | unlock: | |
5955102c | 1723 | inode_unlock(inode); |
f66fd9f0 | 1724 | ceph_free_cap_flush(prealloc_cf); |
ad7a60de LW |
1725 | return ret; |
1726 | } | |
1727 | ||
124e68e7 SW |
1728 | const struct file_operations ceph_file_fops = { |
1729 | .open = ceph_open, | |
1730 | .release = ceph_release, | |
1731 | .llseek = ceph_llseek, | |
3644424d | 1732 | .read_iter = ceph_read_iter, |
4908b822 | 1733 | .write_iter = ceph_write_iter, |
124e68e7 SW |
1734 | .mmap = ceph_mmap, |
1735 | .fsync = ceph_fsync, | |
40819f6f GF |
1736 | .lock = ceph_lock, |
1737 | .flock = ceph_flock, | |
7ce469a5 | 1738 | .splice_read = generic_file_splice_read, |
3551dd79 | 1739 | .splice_write = iter_file_splice_write, |
124e68e7 SW |
1740 | .unlocked_ioctl = ceph_ioctl, |
1741 | .compat_ioctl = ceph_ioctl, | |
ad7a60de | 1742 | .fallocate = ceph_fallocate, |
124e68e7 SW |
1743 | }; |
1744 |