]>
Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
124e68e7 | 2 | |
3d14c5d2 | 3 | #include <linux/module.h> |
124e68e7 | 4 | #include <linux/sched.h> |
5a0e3ad6 | 5 | #include <linux/slab.h> |
124e68e7 | 6 | #include <linux/file.h> |
5ef50c3b | 7 | #include <linux/mount.h> |
124e68e7 SW |
8 | #include <linux/namei.h> |
9 | #include <linux/writeback.h> | |
ad7a60de | 10 | #include <linux/falloc.h> |
124e68e7 SW |
11 | |
12 | #include "super.h" | |
13 | #include "mds_client.h" | |
99ccbd22 | 14 | #include "cache.h" |
124e68e7 SW |
15 | |
16 | /* | |
17 | * Ceph file operations | |
18 | * | |
19 | * Implement basic open/close functionality, and implement | |
20 | * read/write. | |
21 | * | |
22 | * We implement three modes of file I/O: | |
23 | * - buffered uses the generic_file_aio_{read,write} helpers | |
24 | * | |
25 | * - synchronous is used when there is multi-client read/write | |
26 | * sharing, avoids the page cache, and synchronously waits for an | |
27 | * ack from the OSD. | |
28 | * | |
29 | * - direct io takes the variant of the sync path that references | |
30 | * user pages directly. | |
31 | * | |
32 | * fsync() flushes and waits on dirty pages, but just queues metadata | |
33 | * for writeback: since the MDS can recover size and mtime there is no | |
34 | * need to wait for MDS acknowledgement. | |
35 | */ | |
36 | ||
b5b98989 ZC |
37 | /* |
38 | * Calculate the length sum of direct io vectors that can | |
39 | * be combined into one page vector. | |
40 | */ | |
41 | static size_t dio_get_pagev_size(const struct iov_iter *it) | |
42 | { | |
43 | const struct iovec *iov = it->iov; | |
44 | const struct iovec *iovend = iov + it->nr_segs; | |
45 | size_t size; | |
46 | ||
47 | size = iov->iov_len - it->iov_offset; | |
48 | /* | |
49 | * An iov can be page vectored when both the current tail | |
50 | * and the next base are page aligned. | |
51 | */ | |
52 | while (PAGE_ALIGNED((iov->iov_base + iov->iov_len)) && | |
53 | (++iov < iovend && PAGE_ALIGNED((iov->iov_base)))) { | |
54 | size += iov->iov_len; | |
55 | } | |
56 | dout("dio_get_pagevlen len = %zu\n", size); | |
57 | return size; | |
58 | } | |
59 | ||
60 | /* | |
61 | * Allocate a page vector based on (@it, @nbytes). | |
62 | * The return value is the tuple describing a page vector, | |
63 | * that is (@pages, @page_align, @num_pages). | |
64 | */ | |
65 | static struct page ** | |
66 | dio_get_pages_alloc(const struct iov_iter *it, size_t nbytes, | |
67 | size_t *page_align, int *num_pages) | |
68 | { | |
69 | struct iov_iter tmp_it = *it; | |
70 | size_t align; | |
71 | struct page **pages; | |
72 | int ret = 0, idx, npages; | |
73 | ||
74 | align = (unsigned long)(it->iov->iov_base + it->iov_offset) & | |
75 | (PAGE_SIZE - 1); | |
76 | npages = calc_pages_for(align, nbytes); | |
77 | pages = kmalloc(sizeof(*pages) * npages, GFP_KERNEL); | |
78 | if (!pages) { | |
79 | pages = vmalloc(sizeof(*pages) * npages); | |
80 | if (!pages) | |
81 | return ERR_PTR(-ENOMEM); | |
82 | } | |
83 | ||
84 | for (idx = 0; idx < npages; ) { | |
85 | size_t start; | |
86 | ret = iov_iter_get_pages(&tmp_it, pages + idx, nbytes, | |
87 | npages - idx, &start); | |
88 | if (ret < 0) | |
89 | goto fail; | |
90 | ||
91 | iov_iter_advance(&tmp_it, ret); | |
92 | nbytes -= ret; | |
93 | idx += (ret + start + PAGE_SIZE - 1) / PAGE_SIZE; | |
94 | } | |
95 | ||
96 | BUG_ON(nbytes != 0); | |
97 | *num_pages = npages; | |
98 | *page_align = align; | |
99 | dout("dio_get_pages_alloc: got %d pages align %zu\n", npages, align); | |
100 | return pages; | |
101 | fail: | |
102 | ceph_put_page_vector(pages, idx, false); | |
103 | return ERR_PTR(ret); | |
104 | } | |
124e68e7 SW |
105 | |
106 | /* | |
107 | * Prepare an open request. Preallocate ceph_cap to avoid an | |
108 | * inopportune ENOMEM later. | |
109 | */ | |
110 | static struct ceph_mds_request * | |
111 | prepare_open_request(struct super_block *sb, int flags, int create_mode) | |
112 | { | |
3d14c5d2 YS |
113 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
114 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
115 | struct ceph_mds_request *req; |
116 | int want_auth = USE_ANY_MDS; | |
117 | int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN; | |
118 | ||
119 | if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC)) | |
120 | want_auth = USE_AUTH_MDS; | |
121 | ||
122 | req = ceph_mdsc_create_request(mdsc, op, want_auth); | |
123 | if (IS_ERR(req)) | |
124 | goto out; | |
125 | req->r_fmode = ceph_flags_to_mode(flags); | |
126 | req->r_args.open.flags = cpu_to_le32(flags); | |
127 | req->r_args.open.mode = cpu_to_le32(create_mode); | |
124e68e7 SW |
128 | out: |
129 | return req; | |
130 | } | |
131 | ||
132 | /* | |
133 | * initialize private struct file data. | |
134 | * if we fail, clean up by dropping fmode reference on the ceph_inode | |
135 | */ | |
136 | static int ceph_init_file(struct inode *inode, struct file *file, int fmode) | |
137 | { | |
138 | struct ceph_file_info *cf; | |
139 | int ret = 0; | |
140 | ||
141 | switch (inode->i_mode & S_IFMT) { | |
142 | case S_IFREG: | |
46b59b2b YZ |
143 | ceph_fscache_register_inode_cookie(inode); |
144 | ceph_fscache_file_set_cookie(inode, file); | |
124e68e7 SW |
145 | case S_IFDIR: |
146 | dout("init_file %p %p 0%o (regular)\n", inode, file, | |
147 | inode->i_mode); | |
99ec2697 | 148 | cf = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL); |
124e68e7 SW |
149 | if (cf == NULL) { |
150 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
151 | return -ENOMEM; | |
152 | } | |
153 | cf->fmode = fmode; | |
154 | cf->next_offset = 2; | |
fdd4e158 | 155 | cf->readdir_cache_idx = -1; |
124e68e7 SW |
156 | file->private_data = cf; |
157 | BUG_ON(inode->i_fop->release != ceph_release); | |
158 | break; | |
159 | ||
160 | case S_IFLNK: | |
161 | dout("init_file %p %p 0%o (symlink)\n", inode, file, | |
162 | inode->i_mode); | |
163 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
164 | break; | |
165 | ||
166 | default: | |
167 | dout("init_file %p %p 0%o (special)\n", inode, file, | |
168 | inode->i_mode); | |
169 | /* | |
170 | * we need to drop the open ref now, since we don't | |
171 | * have .release set to ceph_release. | |
172 | */ | |
173 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
174 | BUG_ON(inode->i_fop->release == ceph_release); | |
175 | ||
176 | /* call the proper open fop */ | |
177 | ret = inode->i_fop->open(inode, file); | |
178 | } | |
179 | return ret; | |
180 | } | |
181 | ||
77310320 YZ |
182 | /* |
183 | * try renew caps after session gets killed. | |
184 | */ | |
185 | int ceph_renew_caps(struct inode *inode) | |
186 | { | |
187 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; | |
188 | struct ceph_inode_info *ci = ceph_inode(inode); | |
189 | struct ceph_mds_request *req; | |
190 | int err, flags, wanted; | |
191 | ||
192 | spin_lock(&ci->i_ceph_lock); | |
193 | wanted = __ceph_caps_file_wanted(ci); | |
194 | if (__ceph_is_any_real_caps(ci) && | |
195 | (!(wanted & CEPH_CAP_ANY_WR) == 0 || ci->i_auth_cap)) { | |
196 | int issued = __ceph_caps_issued(ci, NULL); | |
197 | spin_unlock(&ci->i_ceph_lock); | |
198 | dout("renew caps %p want %s issued %s updating mds_wanted\n", | |
199 | inode, ceph_cap_string(wanted), ceph_cap_string(issued)); | |
200 | ceph_check_caps(ci, 0, NULL); | |
201 | return 0; | |
202 | } | |
203 | spin_unlock(&ci->i_ceph_lock); | |
204 | ||
205 | flags = 0; | |
206 | if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR)) | |
207 | flags = O_RDWR; | |
208 | else if (wanted & CEPH_CAP_FILE_RD) | |
209 | flags = O_RDONLY; | |
210 | else if (wanted & CEPH_CAP_FILE_WR) | |
211 | flags = O_WRONLY; | |
212 | #ifdef O_LAZY | |
213 | if (wanted & CEPH_CAP_FILE_LAZYIO) | |
214 | flags |= O_LAZY; | |
215 | #endif | |
216 | ||
217 | req = prepare_open_request(inode->i_sb, flags, 0); | |
218 | if (IS_ERR(req)) { | |
219 | err = PTR_ERR(req); | |
220 | goto out; | |
221 | } | |
222 | ||
223 | req->r_inode = inode; | |
224 | ihold(inode); | |
225 | req->r_num_caps = 1; | |
226 | req->r_fmode = -1; | |
227 | ||
228 | err = ceph_mdsc_do_request(mdsc, NULL, req); | |
229 | ceph_mdsc_put_request(req); | |
230 | out: | |
231 | dout("renew caps %p open result=%d\n", inode, err); | |
232 | return err < 0 ? err : 0; | |
233 | } | |
234 | ||
124e68e7 | 235 | /* |
124e68e7 SW |
236 | * If we already have the requisite capabilities, we can satisfy |
237 | * the open request locally (no need to request new caps from the | |
238 | * MDS). We do, however, need to inform the MDS (asynchronously) | |
239 | * if our wanted caps set expands. | |
240 | */ | |
241 | int ceph_open(struct inode *inode, struct file *file) | |
242 | { | |
243 | struct ceph_inode_info *ci = ceph_inode(inode); | |
3d14c5d2 YS |
244 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
245 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
246 | struct ceph_mds_request *req; |
247 | struct ceph_file_info *cf = file->private_data; | |
124e68e7 SW |
248 | int err; |
249 | int flags, fmode, wanted; | |
250 | ||
251 | if (cf) { | |
252 | dout("open file %p is already opened\n", file); | |
253 | return 0; | |
254 | } | |
255 | ||
256 | /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */ | |
257 | flags = file->f_flags & ~(O_CREAT|O_EXCL); | |
258 | if (S_ISDIR(inode->i_mode)) | |
259 | flags = O_DIRECTORY; /* mds likes to know */ | |
260 | ||
261 | dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode, | |
262 | ceph_vinop(inode), file, flags, file->f_flags); | |
263 | fmode = ceph_flags_to_mode(flags); | |
264 | wanted = ceph_caps_for_mode(fmode); | |
265 | ||
266 | /* snapped files are read-only */ | |
267 | if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE)) | |
268 | return -EROFS; | |
269 | ||
270 | /* trivially open snapdir */ | |
271 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
be655596 | 272 | spin_lock(&ci->i_ceph_lock); |
124e68e7 | 273 | __ceph_get_fmode(ci, fmode); |
be655596 | 274 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
275 | return ceph_init_file(inode, file, fmode); |
276 | } | |
277 | ||
278 | /* | |
7421ab80 SW |
279 | * No need to block if we have caps on the auth MDS (for |
280 | * write) or any MDS (for read). Update wanted set | |
124e68e7 SW |
281 | * asynchronously. |
282 | */ | |
be655596 | 283 | spin_lock(&ci->i_ceph_lock); |
7421ab80 SW |
284 | if (__ceph_is_any_real_caps(ci) && |
285 | (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) { | |
124e68e7 SW |
286 | int mds_wanted = __ceph_caps_mds_wanted(ci); |
287 | int issued = __ceph_caps_issued(ci, NULL); | |
288 | ||
289 | dout("open %p fmode %d want %s issued %s using existing\n", | |
290 | inode, fmode, ceph_cap_string(wanted), | |
291 | ceph_cap_string(issued)); | |
292 | __ceph_get_fmode(ci, fmode); | |
be655596 | 293 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
294 | |
295 | /* adjust wanted? */ | |
296 | if ((issued & wanted) != wanted && | |
297 | (mds_wanted & wanted) != wanted && | |
298 | ceph_snap(inode) != CEPH_SNAPDIR) | |
299 | ceph_check_caps(ci, 0, NULL); | |
300 | ||
301 | return ceph_init_file(inode, file, fmode); | |
302 | } else if (ceph_snap(inode) != CEPH_NOSNAP && | |
303 | (ci->i_snap_caps & wanted) == wanted) { | |
304 | __ceph_get_fmode(ci, fmode); | |
be655596 | 305 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
306 | return ceph_init_file(inode, file, fmode); |
307 | } | |
99ccbd22 | 308 | |
be655596 | 309 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
310 | |
311 | dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted)); | |
312 | req = prepare_open_request(inode->i_sb, flags, 0); | |
313 | if (IS_ERR(req)) { | |
314 | err = PTR_ERR(req); | |
315 | goto out; | |
316 | } | |
70b666c3 SW |
317 | req->r_inode = inode; |
318 | ihold(inode); | |
99ccbd22 | 319 | |
124e68e7 | 320 | req->r_num_caps = 1; |
e36d571d | 321 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
124e68e7 SW |
322 | if (!err) |
323 | err = ceph_init_file(inode, file, req->r_fmode); | |
324 | ceph_mdsc_put_request(req); | |
325 | dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode)); | |
326 | out: | |
327 | return err; | |
328 | } | |
329 | ||
330 | ||
331 | /* | |
5ef50c3b SW |
332 | * Do a lookup + open with a single request. If we get a non-existent |
333 | * file or symlink, return 1 so the VFS can retry. | |
124e68e7 | 334 | */ |
5ef50c3b | 335 | int ceph_atomic_open(struct inode *dir, struct dentry *dentry, |
30d90494 | 336 | struct file *file, unsigned flags, umode_t mode, |
d9585277 | 337 | int *opened) |
124e68e7 | 338 | { |
3d14c5d2 YS |
339 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
340 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 | 341 | struct ceph_mds_request *req; |
5ef50c3b | 342 | struct dentry *dn; |
b1ee94aa | 343 | struct ceph_acls_info acls = {}; |
315f2408 | 344 | int mask; |
124e68e7 | 345 | int err; |
124e68e7 | 346 | |
a455589f AV |
347 | dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n", |
348 | dir, dentry, dentry, | |
5ef50c3b SW |
349 | d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode); |
350 | ||
351 | if (dentry->d_name.len > NAME_MAX) | |
352 | return -ENAMETOOLONG; | |
353 | ||
b1ee94aa YZ |
354 | if (flags & O_CREAT) { |
355 | err = ceph_pre_init_acls(dir, &mode, &acls); | |
356 | if (err < 0) | |
357 | return err; | |
358 | } | |
359 | ||
124e68e7 SW |
360 | /* do the open */ |
361 | req = prepare_open_request(dir->i_sb, flags, mode); | |
b1ee94aa YZ |
362 | if (IS_ERR(req)) { |
363 | err = PTR_ERR(req); | |
364 | goto out_acl; | |
365 | } | |
124e68e7 SW |
366 | req->r_dentry = dget(dentry); |
367 | req->r_num_caps = 2; | |
368 | if (flags & O_CREAT) { | |
369 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
370 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
b1ee94aa YZ |
371 | if (acls.pagelist) { |
372 | req->r_pagelist = acls.pagelist; | |
373 | acls.pagelist = NULL; | |
374 | } | |
124e68e7 | 375 | } |
315f2408 YZ |
376 | |
377 | mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED; | |
378 | if (ceph_security_xattr_wanted(dir)) | |
379 | mask |= CEPH_CAP_XATTR_SHARED; | |
380 | req->r_args.open.mask = cpu_to_le32(mask); | |
381 | ||
124e68e7 | 382 | req->r_locked_dir = dir; /* caller holds dir->i_mutex */ |
acda7657 SW |
383 | err = ceph_mdsc_do_request(mdsc, |
384 | (flags & (O_CREAT|O_TRUNC)) ? dir : NULL, | |
385 | req); | |
bf91c315 | 386 | err = ceph_handle_snapdir(req, dentry, err); |
79aec984 | 387 | if (err) |
b1ee94aa | 388 | goto out_req; |
79aec984 | 389 | |
a43137f7 | 390 | if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry) |
124e68e7 | 391 | err = ceph_handle_notrace_create(dir, dentry); |
2d83bde9 | 392 | |
00699ad8 | 393 | if (d_in_lookup(dentry)) { |
5ef50c3b SW |
394 | dn = ceph_finish_lookup(req, dentry, err); |
395 | if (IS_ERR(dn)) | |
396 | err = PTR_ERR(dn); | |
397 | } else { | |
398 | /* we were given a hashed negative dentry */ | |
399 | dn = NULL; | |
400 | } | |
401 | if (err) | |
b1ee94aa | 402 | goto out_req; |
2b0143b5 | 403 | if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) { |
5ef50c3b SW |
404 | /* make vfs retry on splice, ENOENT, or symlink */ |
405 | dout("atomic_open finish_no_open on dn %p\n", dn); | |
406 | err = finish_no_open(file, dn); | |
407 | } else { | |
408 | dout("atomic_open finish_open on dn %p\n", dn); | |
6e8575fa | 409 | if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) { |
2b0143b5 | 410 | ceph_init_inode_acls(d_inode(dentry), &acls); |
6e8575fa SL |
411 | *opened |= FILE_CREATED; |
412 | } | |
5ef50c3b SW |
413 | err = finish_open(file, dentry, ceph_open, opened); |
414 | } | |
b1ee94aa | 415 | out_req: |
ab866549 YZ |
416 | if (!req->r_err && req->r_target_inode) |
417 | ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode); | |
5ef50c3b | 418 | ceph_mdsc_put_request(req); |
b1ee94aa YZ |
419 | out_acl: |
420 | ceph_release_acls_info(&acls); | |
5ef50c3b | 421 | dout("atomic_open result=%d\n", err); |
d9585277 | 422 | return err; |
124e68e7 SW |
423 | } |
424 | ||
425 | int ceph_release(struct inode *inode, struct file *file) | |
426 | { | |
427 | struct ceph_inode_info *ci = ceph_inode(inode); | |
428 | struct ceph_file_info *cf = file->private_data; | |
429 | ||
430 | dout("release inode %p file %p\n", inode, file); | |
431 | ceph_put_fmode(ci, cf->fmode); | |
432 | if (cf->last_readdir) | |
433 | ceph_mdsc_put_request(cf->last_readdir); | |
434 | kfree(cf->last_name); | |
435 | kfree(cf->dir_info); | |
124e68e7 | 436 | kmem_cache_free(ceph_file_cachep, cf); |
195d3ce2 SW |
437 | |
438 | /* wake up anyone waiting for caps on this inode */ | |
03066f23 | 439 | wake_up_all(&ci->i_cap_wq); |
124e68e7 SW |
440 | return 0; |
441 | } | |
442 | ||
83701246 | 443 | enum { |
c8fe9b17 YZ |
444 | HAVE_RETRIED = 1, |
445 | CHECK_EOF = 2, | |
446 | READ_INLINE = 3, | |
83701246 YZ |
447 | }; |
448 | ||
124e68e7 SW |
449 | /* |
450 | * Read a range of bytes striped over one or more objects. Iterate over | |
451 | * objects we stripe over. (That's not atomic, but good enough for now.) | |
452 | * | |
453 | * If we get a short result from the OSD, check against i_size; we need to | |
454 | * only return a short read to the caller if we hit EOF. | |
455 | */ | |
456 | static int striped_read(struct inode *inode, | |
7ce469a5 | 457 | u64 pos, u64 len, |
6a026589 | 458 | struct page **pages, int num_pages, |
7ce469a5 | 459 | int page_align, int *checkeof) |
124e68e7 | 460 | { |
3d14c5d2 | 461 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
124e68e7 | 462 | struct ceph_inode_info *ci = ceph_inode(inode); |
7ce469a5 | 463 | u64 this_len; |
99c88e69 | 464 | loff_t i_size; |
7ce469a5 YZ |
465 | int page_idx; |
466 | int ret, read = 0; | |
124e68e7 SW |
467 | bool hit_stripe, was_short; |
468 | ||
469 | /* | |
470 | * we may need to do multiple reads. not atomic, unfortunately. | |
471 | */ | |
124e68e7 | 472 | more: |
7ce469a5 YZ |
473 | this_len = len; |
474 | page_idx = (page_align + read) >> PAGE_SHIFT; | |
3d14c5d2 | 475 | ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), |
124e68e7 | 476 | &ci->i_layout, pos, &this_len, |
7ce469a5 YZ |
477 | ci->i_truncate_seq, ci->i_truncate_size, |
478 | pages + page_idx, num_pages - page_idx, | |
479 | ((page_align + read) & ~PAGE_MASK)); | |
124e68e7 SW |
480 | if (ret == -ENOENT) |
481 | ret = 0; | |
7ce469a5 | 482 | hit_stripe = this_len < len; |
0e98728f | 483 | was_short = ret >= 0 && ret < this_len; |
7ce469a5 | 484 | dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, len, read, |
124e68e7 SW |
485 | ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : ""); |
486 | ||
99c88e69 | 487 | i_size = i_size_read(inode); |
02ae66d8 | 488 | if (ret >= 0) { |
99c88e69 YZ |
489 | if (was_short && (pos + ret < i_size)) { |
490 | int zlen = min(this_len - ret, i_size - pos - ret); | |
7ce469a5 | 491 | int zoff = page_align + read + ret; |
02ae66d8 | 492 | dout(" zero gap %llu to %llu\n", |
7ce469a5 | 493 | pos + ret, pos + ret + zlen); |
1487a688 YZ |
494 | ceph_zero_page_vector_range(zoff, zlen, pages); |
495 | ret += zlen; | |
124e68e7 | 496 | } |
02ae66d8 | 497 | |
7ce469a5 | 498 | read += ret; |
124e68e7 | 499 | pos += ret; |
7ce469a5 | 500 | len -= ret; |
124e68e7 | 501 | |
02ae66d8 | 502 | /* hit stripe and need continue*/ |
7ce469a5 | 503 | if (len && hit_stripe && pos < i_size) |
124e68e7 SW |
504 | goto more; |
505 | } | |
506 | ||
ee7289bf | 507 | if (read > 0) { |
02ae66d8 | 508 | ret = read; |
c3cd6283 | 509 | /* did we bounce off eof? */ |
7ce469a5 | 510 | if (pos + len > i_size) |
83701246 | 511 | *checkeof = CHECK_EOF; |
124e68e7 SW |
512 | } |
513 | ||
124e68e7 SW |
514 | dout("striped_read returns %d\n", ret); |
515 | return ret; | |
516 | } | |
517 | ||
518 | /* | |
519 | * Completely synchronous read and write methods. Direct from __user | |
520 | * buffer to osd, or directly to user pages (if O_DIRECT). | |
521 | * | |
522 | * If the read spans object boundary, just do multiple reads. | |
523 | */ | |
7ce469a5 YZ |
524 | static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to, |
525 | int *checkeof) | |
124e68e7 | 526 | { |
8eb4efb0 | 527 | struct file *file = iocb->ki_filp; |
496ad9aa | 528 | struct inode *inode = file_inode(file); |
124e68e7 | 529 | struct page **pages; |
8eb4efb0 | 530 | u64 off = iocb->ki_pos; |
7ce469a5 YZ |
531 | int num_pages; |
532 | ssize_t ret; | |
533 | size_t len = iov_iter_count(to); | |
124e68e7 | 534 | |
8eb4efb0 | 535 | dout("sync_read on file %p %llu~%u %s\n", file, off, |
536 | (unsigned)len, | |
124e68e7 | 537 | (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); |
d0d0db22 YZ |
538 | |
539 | if (!len) | |
540 | return 0; | |
e98b6fed SW |
541 | /* |
542 | * flush any page cache pages in this range. this | |
543 | * will make concurrent normal and sync io slow, | |
544 | * but it will at least behave sensibly when they are | |
545 | * in sequence. | |
546 | */ | |
8eb4efb0 | 547 | ret = filemap_write_and_wait_range(inode->i_mapping, off, |
548 | off + len); | |
29065a51 | 549 | if (ret < 0) |
8eb4efb0 | 550 | return ret; |
29065a51 | 551 | |
7ce469a5 YZ |
552 | if (unlikely(to->type & ITER_PIPE)) { |
553 | size_t page_off; | |
554 | ret = iov_iter_get_pages_alloc(to, &pages, len, | |
555 | &page_off); | |
556 | if (ret <= 0) | |
557 | return -ENOMEM; | |
558 | num_pages = DIV_ROUND_UP(ret + page_off, PAGE_SIZE); | |
559 | ||
560 | ret = striped_read(inode, off, ret, pages, num_pages, | |
561 | page_off, checkeof); | |
562 | if (ret > 0) { | |
563 | iov_iter_advance(to, ret); | |
564 | off += ret; | |
565 | } else { | |
566 | iov_iter_advance(to, 0); | |
567 | } | |
568 | ceph_put_page_vector(pages, num_pages, false); | |
569 | } else { | |
570 | num_pages = calc_pages_for(off, len); | |
571 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); | |
572 | if (IS_ERR(pages)) | |
573 | return PTR_ERR(pages); | |
574 | ||
575 | ret = striped_read(inode, off, len, pages, num_pages, | |
576 | (off & ~PAGE_MASK), checkeof); | |
577 | if (ret > 0) { | |
578 | int l, k = 0; | |
579 | size_t left = ret; | |
580 | ||
581 | while (left) { | |
582 | size_t page_off = off & ~PAGE_MASK; | |
583 | size_t copy = min_t(size_t, left, | |
584 | PAGE_SIZE - page_off); | |
585 | l = copy_page_to_iter(pages[k++], page_off, | |
586 | copy, to); | |
587 | off += l; | |
588 | left -= l; | |
589 | if (l < copy) | |
590 | break; | |
591 | } | |
8eb4efb0 | 592 | } |
7ce469a5 | 593 | ceph_release_page_vector(pages, num_pages); |
8eb4efb0 | 594 | } |
124e68e7 | 595 | |
8eb4efb0 | 596 | if (off > iocb->ki_pos) { |
597 | ret = off - iocb->ki_pos; | |
598 | iocb->ki_pos = off; | |
599 | } | |
124e68e7 | 600 | |
7ce469a5 | 601 | dout("sync_read result %zd\n", ret); |
124e68e7 SW |
602 | return ret; |
603 | } | |
604 | ||
c8fe9b17 YZ |
605 | struct ceph_aio_request { |
606 | struct kiocb *iocb; | |
607 | size_t total_len; | |
608 | int write; | |
609 | int error; | |
610 | struct list_head osd_reqs; | |
611 | unsigned num_reqs; | |
612 | atomic_t pending_reqs; | |
5be0389d | 613 | struct timespec mtime; |
c8fe9b17 YZ |
614 | struct ceph_cap_flush *prealloc_cf; |
615 | }; | |
616 | ||
5be0389d YZ |
617 | struct ceph_aio_work { |
618 | struct work_struct work; | |
619 | struct ceph_osd_request *req; | |
620 | }; | |
621 | ||
622 | static void ceph_aio_retry_work(struct work_struct *work); | |
623 | ||
c8fe9b17 YZ |
624 | static void ceph_aio_complete(struct inode *inode, |
625 | struct ceph_aio_request *aio_req) | |
626 | { | |
627 | struct ceph_inode_info *ci = ceph_inode(inode); | |
628 | int ret; | |
629 | ||
630 | if (!atomic_dec_and_test(&aio_req->pending_reqs)) | |
631 | return; | |
632 | ||
633 | ret = aio_req->error; | |
634 | if (!ret) | |
635 | ret = aio_req->total_len; | |
636 | ||
637 | dout("ceph_aio_complete %p rc %d\n", inode, ret); | |
638 | ||
639 | if (ret >= 0 && aio_req->write) { | |
640 | int dirty; | |
641 | ||
642 | loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len; | |
643 | if (endoff > i_size_read(inode)) { | |
644 | if (ceph_inode_set_size(inode, endoff)) | |
645 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); | |
646 | } | |
647 | ||
648 | spin_lock(&ci->i_ceph_lock); | |
649 | ci->i_inline_version = CEPH_INLINE_NONE; | |
650 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, | |
651 | &aio_req->prealloc_cf); | |
652 | spin_unlock(&ci->i_ceph_lock); | |
653 | if (dirty) | |
654 | __mark_inode_dirty(inode, dirty); | |
655 | ||
656 | } | |
657 | ||
658 | ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR : | |
659 | CEPH_CAP_FILE_RD)); | |
660 | ||
661 | aio_req->iocb->ki_complete(aio_req->iocb, ret, 0); | |
662 | ||
663 | ceph_free_cap_flush(aio_req->prealloc_cf); | |
664 | kfree(aio_req); | |
665 | } | |
666 | ||
85e084fe | 667 | static void ceph_aio_complete_req(struct ceph_osd_request *req) |
c8fe9b17 YZ |
668 | { |
669 | int rc = req->r_result; | |
670 | struct inode *inode = req->r_inode; | |
671 | struct ceph_aio_request *aio_req = req->r_priv; | |
672 | struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0); | |
673 | int num_pages = calc_pages_for((u64)osd_data->alignment, | |
674 | osd_data->length); | |
675 | ||
676 | dout("ceph_aio_complete_req %p rc %d bytes %llu\n", | |
677 | inode, rc, osd_data->length); | |
678 | ||
679 | if (rc == -EOLDSNAPC) { | |
5be0389d YZ |
680 | struct ceph_aio_work *aio_work; |
681 | BUG_ON(!aio_req->write); | |
682 | ||
683 | aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS); | |
684 | if (aio_work) { | |
685 | INIT_WORK(&aio_work->work, ceph_aio_retry_work); | |
686 | aio_work->req = req; | |
687 | queue_work(ceph_inode_to_client(inode)->wb_wq, | |
688 | &aio_work->work); | |
689 | return; | |
690 | } | |
691 | rc = -ENOMEM; | |
692 | } else if (!aio_req->write) { | |
c8fe9b17 YZ |
693 | if (rc == -ENOENT) |
694 | rc = 0; | |
695 | if (rc >= 0 && osd_data->length > rc) { | |
696 | int zoff = osd_data->alignment + rc; | |
697 | int zlen = osd_data->length - rc; | |
698 | /* | |
699 | * If read is satisfied by single OSD request, | |
700 | * it can pass EOF. Otherwise read is within | |
701 | * i_size. | |
702 | */ | |
703 | if (aio_req->num_reqs == 1) { | |
704 | loff_t i_size = i_size_read(inode); | |
705 | loff_t endoff = aio_req->iocb->ki_pos + rc; | |
706 | if (endoff < i_size) | |
707 | zlen = min_t(size_t, zlen, | |
708 | i_size - endoff); | |
709 | aio_req->total_len = rc + zlen; | |
710 | } | |
711 | ||
712 | if (zlen > 0) | |
713 | ceph_zero_page_vector_range(zoff, zlen, | |
714 | osd_data->pages); | |
715 | } | |
716 | } | |
717 | ||
a22bd5ff | 718 | ceph_put_page_vector(osd_data->pages, num_pages, !aio_req->write); |
c8fe9b17 YZ |
719 | ceph_osdc_put_request(req); |
720 | ||
721 | if (rc < 0) | |
722 | cmpxchg(&aio_req->error, 0, rc); | |
723 | ||
724 | ceph_aio_complete(inode, aio_req); | |
725 | return; | |
726 | } | |
727 | ||
5be0389d YZ |
728 | static void ceph_aio_retry_work(struct work_struct *work) |
729 | { | |
730 | struct ceph_aio_work *aio_work = | |
731 | container_of(work, struct ceph_aio_work, work); | |
732 | struct ceph_osd_request *orig_req = aio_work->req; | |
733 | struct ceph_aio_request *aio_req = orig_req->r_priv; | |
734 | struct inode *inode = orig_req->r_inode; | |
735 | struct ceph_inode_info *ci = ceph_inode(inode); | |
736 | struct ceph_snap_context *snapc; | |
737 | struct ceph_osd_request *req; | |
738 | int ret; | |
739 | ||
740 | spin_lock(&ci->i_ceph_lock); | |
741 | if (__ceph_have_pending_cap_snap(ci)) { | |
742 | struct ceph_cap_snap *capsnap = | |
743 | list_last_entry(&ci->i_cap_snaps, | |
744 | struct ceph_cap_snap, | |
745 | ci_item); | |
746 | snapc = ceph_get_snap_context(capsnap->context); | |
747 | } else { | |
748 | BUG_ON(!ci->i_head_snapc); | |
749 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
750 | } | |
751 | spin_unlock(&ci->i_ceph_lock); | |
752 | ||
753 | req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 2, | |
754 | false, GFP_NOFS); | |
1418bf07 DC |
755 | if (!req) { |
756 | ret = -ENOMEM; | |
5be0389d YZ |
757 | req = orig_req; |
758 | goto out; | |
759 | } | |
760 | ||
761 | req->r_flags = CEPH_OSD_FLAG_ORDERSNAP | | |
762 | CEPH_OSD_FLAG_ONDISK | | |
763 | CEPH_OSD_FLAG_WRITE; | |
63244fa1 | 764 | ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc); |
d30291b9 | 765 | ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid); |
5be0389d | 766 | |
13d1ad16 ID |
767 | ret = ceph_osdc_alloc_messages(req, GFP_NOFS); |
768 | if (ret) { | |
769 | ceph_osdc_put_request(req); | |
770 | req = orig_req; | |
771 | goto out; | |
772 | } | |
5be0389d YZ |
773 | |
774 | req->r_ops[0] = orig_req->r_ops[0]; | |
775 | osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0); | |
776 | ||
bb873b53 ID |
777 | req->r_mtime = aio_req->mtime; |
778 | req->r_data_offset = req->r_ops[0].extent.offset; | |
5be0389d | 779 | |
5be0389d YZ |
780 | ceph_osdc_put_request(orig_req); |
781 | ||
782 | req->r_callback = ceph_aio_complete_req; | |
783 | req->r_inode = inode; | |
784 | req->r_priv = aio_req; | |
785 | ||
786 | ret = ceph_osdc_start_request(req->r_osdc, req, false); | |
787 | out: | |
788 | if (ret < 0) { | |
5be0389d | 789 | req->r_result = ret; |
85e084fe | 790 | ceph_aio_complete_req(req); |
5be0389d YZ |
791 | } |
792 | ||
db6aed70 | 793 | ceph_put_snap_context(snapc); |
5be0389d YZ |
794 | kfree(aio_work); |
795 | } | |
796 | ||
124e68e7 | 797 | /* |
26be8808 AE |
798 | * Write commit request unsafe callback, called to tell us when a |
799 | * request is unsafe (that is, in flight--has been handed to the | |
800 | * messenger to send to its target osd). It is called again when | |
801 | * we've received a response message indicating the request is | |
802 | * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request | |
803 | * is completed early (and unsuccessfully) due to a timeout or | |
804 | * interrupt. | |
805 | * | |
806 | * This is used if we requested both an ACK and ONDISK commit reply | |
807 | * from the OSD. | |
124e68e7 | 808 | */ |
26be8808 | 809 | static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe) |
124e68e7 SW |
810 | { |
811 | struct ceph_inode_info *ci = ceph_inode(req->r_inode); | |
812 | ||
26be8808 AE |
813 | dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid, |
814 | unsafe ? "un" : ""); | |
815 | if (unsafe) { | |
816 | ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR); | |
817 | spin_lock(&ci->i_unsafe_lock); | |
818 | list_add_tail(&req->r_unsafe_item, | |
819 | &ci->i_unsafe_writes); | |
820 | spin_unlock(&ci->i_unsafe_lock); | |
fe5da05e ID |
821 | |
822 | complete_all(&req->r_completion); | |
26be8808 AE |
823 | } else { |
824 | spin_lock(&ci->i_unsafe_lock); | |
825 | list_del_init(&req->r_unsafe_item); | |
826 | spin_unlock(&ci->i_unsafe_lock); | |
827 | ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR); | |
828 | } | |
124e68e7 SW |
829 | } |
830 | ||
9a5530c6 YZ |
831 | /* |
832 | * Wait on any unsafe replies for the given inode. First wait on the | |
833 | * newest request, and make that the upper bound. Then, if there are | |
834 | * more requests, keep waiting on the oldest as long as it is still older | |
835 | * than the original request. | |
836 | */ | |
837 | void ceph_sync_write_wait(struct inode *inode) | |
838 | { | |
839 | struct ceph_inode_info *ci = ceph_inode(inode); | |
840 | struct list_head *head = &ci->i_unsafe_writes; | |
841 | struct ceph_osd_request *req; | |
842 | u64 last_tid; | |
843 | ||
844 | if (!S_ISREG(inode->i_mode)) | |
845 | return; | |
846 | ||
847 | spin_lock(&ci->i_unsafe_lock); | |
848 | if (list_empty(head)) | |
849 | goto out; | |
850 | ||
851 | /* set upper bound as _last_ entry in chain */ | |
852 | ||
853 | req = list_last_entry(head, struct ceph_osd_request, | |
854 | r_unsafe_item); | |
855 | last_tid = req->r_tid; | |
856 | ||
857 | do { | |
858 | ceph_osdc_get_request(req); | |
859 | spin_unlock(&ci->i_unsafe_lock); | |
860 | ||
861 | dout("sync_write_wait on tid %llu (until %llu)\n", | |
862 | req->r_tid, last_tid); | |
c297eb42 | 863 | wait_for_completion(&req->r_done_completion); |
9a5530c6 YZ |
864 | ceph_osdc_put_request(req); |
865 | ||
866 | spin_lock(&ci->i_unsafe_lock); | |
867 | /* | |
868 | * from here on look at first entry in chain, since we | |
869 | * only want to wait for anything older than last_tid | |
870 | */ | |
871 | if (list_empty(head)) | |
872 | break; | |
873 | req = list_first_entry(head, struct ceph_osd_request, | |
874 | r_unsafe_item); | |
875 | } while (req->r_tid < last_tid); | |
876 | out: | |
877 | spin_unlock(&ci->i_unsafe_lock); | |
878 | } | |
e8344e66 | 879 | |
e8344e66 | 880 | static ssize_t |
c8fe9b17 YZ |
881 | ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter, |
882 | struct ceph_snap_context *snapc, | |
883 | struct ceph_cap_flush **pcf) | |
124e68e7 | 884 | { |
e8344e66 | 885 | struct file *file = iocb->ki_filp; |
496ad9aa | 886 | struct inode *inode = file_inode(file); |
124e68e7 | 887 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 | 888 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
acead002 | 889 | struct ceph_vino vino; |
124e68e7 SW |
890 | struct ceph_osd_request *req; |
891 | struct page **pages; | |
c8fe9b17 YZ |
892 | struct ceph_aio_request *aio_req = NULL; |
893 | int num_pages = 0; | |
124e68e7 | 894 | int flags; |
124e68e7 | 895 | int ret; |
c2050a45 | 896 | struct timespec mtime = current_time(inode); |
c8fe9b17 YZ |
897 | size_t count = iov_iter_count(iter); |
898 | loff_t pos = iocb->ki_pos; | |
899 | bool write = iov_iter_rw(iter) == WRITE; | |
124e68e7 | 900 | |
c8fe9b17 | 901 | if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
124e68e7 SW |
902 | return -EROFS; |
903 | ||
c8fe9b17 YZ |
904 | dout("sync_direct_read_write (%s) on file %p %lld~%u\n", |
905 | (write ? "write" : "read"), file, pos, (unsigned)count); | |
124e68e7 | 906 | |
e8344e66 | 907 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); |
29065a51 YS |
908 | if (ret < 0) |
909 | return ret; | |
910 | ||
c8fe9b17 | 911 | if (write) { |
5d7eb1a3 | 912 | int ret2 = invalidate_inode_pages2_range(inode->i_mapping, |
09cbfeaf KS |
913 | pos >> PAGE_SHIFT, |
914 | (pos + count) >> PAGE_SHIFT); | |
5d7eb1a3 | 915 | if (ret2 < 0) |
a380a031 | 916 | dout("invalidate_inode_pages2_range returned %d\n", ret2); |
29065a51 | 917 | |
c8fe9b17 YZ |
918 | flags = CEPH_OSD_FLAG_ORDERSNAP | |
919 | CEPH_OSD_FLAG_ONDISK | | |
920 | CEPH_OSD_FLAG_WRITE; | |
921 | } else { | |
922 | flags = CEPH_OSD_FLAG_READ; | |
923 | } | |
124e68e7 | 924 | |
c8fe9b17 YZ |
925 | while (iov_iter_count(iter) > 0) { |
926 | u64 size = dio_get_pagev_size(iter); | |
927 | size_t start = 0; | |
928 | ssize_t len; | |
e8344e66 | 929 | |
e8344e66 | 930 | vino = ceph_vino(inode); |
931 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
c8fe9b17 YZ |
932 | vino, pos, &size, 0, |
933 | /*include a 'startsync' command*/ | |
934 | write ? 2 : 1, | |
935 | write ? CEPH_OSD_OP_WRITE : | |
936 | CEPH_OSD_OP_READ, | |
937 | flags, snapc, | |
e8344e66 | 938 | ci->i_truncate_seq, |
939 | ci->i_truncate_size, | |
940 | false); | |
941 | if (IS_ERR(req)) { | |
942 | ret = PTR_ERR(req); | |
eab87235 | 943 | break; |
e8344e66 | 944 | } |
124e68e7 | 945 | |
c8fe9b17 YZ |
946 | len = size; |
947 | pages = dio_get_pages_alloc(iter, len, &start, &num_pages); | |
b5b98989 | 948 | if (IS_ERR(pages)) { |
64c31311 | 949 | ceph_osdc_put_request(req); |
b5b98989 | 950 | ret = PTR_ERR(pages); |
64c31311 | 951 | break; |
124e68e7 SW |
952 | } |
953 | ||
954 | /* | |
c8fe9b17 YZ |
955 | * To simplify error handling, allow AIO when IO within i_size |
956 | * or IO can be satisfied by single OSD request. | |
124e68e7 | 957 | */ |
c8fe9b17 YZ |
958 | if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) && |
959 | (len == count || pos + count <= i_size_read(inode))) { | |
960 | aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL); | |
961 | if (aio_req) { | |
962 | aio_req->iocb = iocb; | |
963 | aio_req->write = write; | |
964 | INIT_LIST_HEAD(&aio_req->osd_reqs); | |
965 | if (write) { | |
5be0389d | 966 | aio_req->mtime = mtime; |
c8fe9b17 YZ |
967 | swap(aio_req->prealloc_cf, *pcf); |
968 | } | |
969 | } | |
970 | /* ignore error */ | |
971 | } | |
972 | ||
973 | if (write) { | |
974 | /* | |
975 | * throw out any page cache pages in this range. this | |
976 | * may block. | |
977 | */ | |
978 | truncate_inode_pages_range(inode->i_mapping, pos, | |
09cbfeaf | 979 | (pos+len) | (PAGE_SIZE - 1)); |
c8fe9b17 YZ |
980 | |
981 | osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0); | |
bb873b53 | 982 | req->r_mtime = mtime; |
c8fe9b17 YZ |
983 | } |
984 | ||
c8fe9b17 YZ |
985 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, start, |
986 | false, false); | |
e8344e66 | 987 | |
c8fe9b17 YZ |
988 | if (aio_req) { |
989 | aio_req->total_len += len; | |
990 | aio_req->num_reqs++; | |
991 | atomic_inc(&aio_req->pending_reqs); | |
992 | ||
993 | req->r_callback = ceph_aio_complete_req; | |
994 | req->r_inode = inode; | |
995 | req->r_priv = aio_req; | |
996 | list_add_tail(&req->r_unsafe_item, &aio_req->osd_reqs); | |
997 | ||
998 | pos += len; | |
999 | iov_iter_advance(iter, len); | |
1000 | continue; | |
1001 | } | |
1002 | ||
1003 | ret = ceph_osdc_start_request(req->r_osdc, req, false); | |
e8344e66 | 1004 | if (!ret) |
1005 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1006 | ||
c8fe9b17 YZ |
1007 | size = i_size_read(inode); |
1008 | if (!write) { | |
1009 | if (ret == -ENOENT) | |
1010 | ret = 0; | |
1011 | if (ret >= 0 && ret < len && pos + ret < size) { | |
1012 | int zlen = min_t(size_t, len - ret, | |
1013 | size - pos - ret); | |
1014 | ceph_zero_page_vector_range(start + ret, zlen, | |
1015 | pages); | |
1016 | ret += zlen; | |
1017 | } | |
1018 | if (ret >= 0) | |
1019 | len = ret; | |
1020 | } | |
1021 | ||
a22bd5ff | 1022 | ceph_put_page_vector(pages, num_pages, !write); |
e8344e66 | 1023 | |
e8344e66 | 1024 | ceph_osdc_put_request(req); |
c8fe9b17 | 1025 | if (ret < 0) |
e8344e66 | 1026 | break; |
64c31311 | 1027 | |
c8fe9b17 YZ |
1028 | pos += len; |
1029 | iov_iter_advance(iter, len); | |
1030 | ||
1031 | if (!write && pos >= size) | |
e8344e66 | 1032 | break; |
64c31311 | 1033 | |
c8fe9b17 YZ |
1034 | if (write && pos > size) { |
1035 | if (ceph_inode_set_size(inode, pos)) | |
64c31311 AV |
1036 | ceph_check_caps(ceph_inode(inode), |
1037 | CHECK_CAPS_AUTHONLY, | |
1038 | NULL); | |
1039 | } | |
e8344e66 | 1040 | } |
1041 | ||
c8fe9b17 | 1042 | if (aio_req) { |
fc8c3892 YZ |
1043 | LIST_HEAD(osd_reqs); |
1044 | ||
c8fe9b17 YZ |
1045 | if (aio_req->num_reqs == 0) { |
1046 | kfree(aio_req); | |
1047 | return ret; | |
1048 | } | |
1049 | ||
1050 | ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR : | |
1051 | CEPH_CAP_FILE_RD); | |
1052 | ||
fc8c3892 YZ |
1053 | list_splice(&aio_req->osd_reqs, &osd_reqs); |
1054 | while (!list_empty(&osd_reqs)) { | |
1055 | req = list_first_entry(&osd_reqs, | |
c8fe9b17 YZ |
1056 | struct ceph_osd_request, |
1057 | r_unsafe_item); | |
1058 | list_del_init(&req->r_unsafe_item); | |
1059 | if (ret >= 0) | |
1060 | ret = ceph_osdc_start_request(req->r_osdc, | |
1061 | req, false); | |
1062 | if (ret < 0) { | |
1063 | req->r_result = ret; | |
85e084fe | 1064 | ceph_aio_complete_req(req); |
c8fe9b17 YZ |
1065 | } |
1066 | } | |
1067 | return -EIOCBQUEUED; | |
1068 | } | |
1069 | ||
1070 | if (ret != -EOLDSNAPC && pos > iocb->ki_pos) { | |
1071 | ret = pos - iocb->ki_pos; | |
e8344e66 | 1072 | iocb->ki_pos = pos; |
e8344e66 | 1073 | } |
1074 | return ret; | |
1075 | } | |
1076 | ||
e8344e66 | 1077 | /* |
1078 | * Synchronous write, straight from __user pointer or user pages. | |
1079 | * | |
1080 | * If write spans object boundary, just do multiple writes. (For a | |
1081 | * correct atomic write, we should e.g. take write locks on all | |
1082 | * objects, rollback on failure, etc.) | |
1083 | */ | |
06fee30f | 1084 | static ssize_t |
5dda377c YZ |
1085 | ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos, |
1086 | struct ceph_snap_context *snapc) | |
e8344e66 | 1087 | { |
1088 | struct file *file = iocb->ki_filp; | |
1089 | struct inode *inode = file_inode(file); | |
1090 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1091 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
e8344e66 | 1092 | struct ceph_vino vino; |
1093 | struct ceph_osd_request *req; | |
1094 | struct page **pages; | |
1095 | u64 len; | |
1096 | int num_pages; | |
1097 | int written = 0; | |
1098 | int flags; | |
1099 | int check_caps = 0; | |
1100 | int ret; | |
c2050a45 | 1101 | struct timespec mtime = current_time(inode); |
4908b822 | 1102 | size_t count = iov_iter_count(from); |
e8344e66 | 1103 | |
1104 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) | |
1105 | return -EROFS; | |
1106 | ||
1107 | dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count); | |
1108 | ||
1109 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); | |
1110 | if (ret < 0) | |
1111 | return ret; | |
1112 | ||
1113 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
09cbfeaf KS |
1114 | pos >> PAGE_SHIFT, |
1115 | (pos + count) >> PAGE_SHIFT); | |
e8344e66 | 1116 | if (ret < 0) |
1117 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
1118 | ||
1119 | flags = CEPH_OSD_FLAG_ORDERSNAP | | |
1120 | CEPH_OSD_FLAG_ONDISK | | |
1121 | CEPH_OSD_FLAG_WRITE | | |
1122 | CEPH_OSD_FLAG_ACK; | |
1123 | ||
4908b822 | 1124 | while ((len = iov_iter_count(from)) > 0) { |
e8344e66 | 1125 | size_t left; |
1126 | int n; | |
1127 | ||
e8344e66 | 1128 | vino = ceph_vino(inode); |
1129 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
715e4cd4 | 1130 | vino, pos, &len, 0, 1, |
e8344e66 | 1131 | CEPH_OSD_OP_WRITE, flags, snapc, |
1132 | ci->i_truncate_seq, | |
1133 | ci->i_truncate_size, | |
1134 | false); | |
1135 | if (IS_ERR(req)) { | |
1136 | ret = PTR_ERR(req); | |
eab87235 | 1137 | break; |
e8344e66 | 1138 | } |
1139 | ||
1140 | /* | |
1141 | * write from beginning of first page, | |
1142 | * regardless of io alignment | |
1143 | */ | |
09cbfeaf | 1144 | num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
e8344e66 | 1145 | |
687265e5 | 1146 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
124e68e7 SW |
1147 | if (IS_ERR(pages)) { |
1148 | ret = PTR_ERR(pages); | |
1149 | goto out; | |
1150 | } | |
e8344e66 | 1151 | |
1152 | left = len; | |
1153 | for (n = 0; n < num_pages; n++) { | |
125d725c | 1154 | size_t plen = min_t(size_t, left, PAGE_SIZE); |
4908b822 | 1155 | ret = copy_page_from_iter(pages[n], 0, plen, from); |
e8344e66 | 1156 | if (ret != plen) { |
1157 | ret = -EFAULT; | |
1158 | break; | |
1159 | } | |
1160 | left -= ret; | |
e8344e66 | 1161 | } |
1162 | ||
124e68e7 SW |
1163 | if (ret < 0) { |
1164 | ceph_release_page_vector(pages, num_pages); | |
1165 | goto out; | |
1166 | } | |
1167 | ||
e8344e66 | 1168 | /* get a second commit callback */ |
1169 | req->r_unsafe_callback = ceph_sync_write_unsafe; | |
1170 | req->r_inode = inode; | |
124e68e7 | 1171 | |
e8344e66 | 1172 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, |
1173 | false, true); | |
02ee07d3 | 1174 | |
bb873b53 | 1175 | req->r_mtime = mtime; |
e8344e66 | 1176 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
1177 | if (!ret) | |
1178 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
124e68e7 SW |
1179 | |
1180 | out: | |
e8344e66 | 1181 | ceph_osdc_put_request(req); |
1182 | if (ret == 0) { | |
1183 | pos += len; | |
1184 | written += len; | |
1185 | ||
1186 | if (pos > i_size_read(inode)) { | |
1187 | check_caps = ceph_inode_set_size(inode, pos); | |
1188 | if (check_caps) | |
1189 | ceph_check_caps(ceph_inode(inode), | |
1190 | CHECK_CAPS_AUTHONLY, | |
1191 | NULL); | |
1192 | } | |
1193 | } else | |
1194 | break; | |
1195 | } | |
124e68e7 | 1196 | |
e8344e66 | 1197 | if (ret != -EOLDSNAPC && written > 0) { |
124e68e7 | 1198 | ret = written; |
e8344e66 | 1199 | iocb->ki_pos = pos; |
124e68e7 SW |
1200 | } |
1201 | return ret; | |
1202 | } | |
1203 | ||
1204 | /* | |
1205 | * Wrap generic_file_aio_read with checks for cap bits on the inode. | |
1206 | * Atomically grab references, so that those bits are not released | |
1207 | * back to the MDS mid-read. | |
1208 | * | |
1209 | * Hmm, the sync read case isn't actually async... should it be? | |
1210 | */ | |
3644424d | 1211 | static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) |
124e68e7 SW |
1212 | { |
1213 | struct file *filp = iocb->ki_filp; | |
2962507c | 1214 | struct ceph_file_info *fi = filp->private_data; |
66ee59af | 1215 | size_t len = iov_iter_count(to); |
496ad9aa | 1216 | struct inode *inode = file_inode(filp); |
124e68e7 | 1217 | struct ceph_inode_info *ci = ceph_inode(inode); |
3738daa6 | 1218 | struct page *pinned_page = NULL; |
124e68e7 | 1219 | ssize_t ret; |
2962507c | 1220 | int want, got = 0; |
83701246 | 1221 | int retry_op = 0, read = 0; |
124e68e7 | 1222 | |
6a026589 | 1223 | again: |
8eb4efb0 | 1224 | dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n", |
1225 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode); | |
1226 | ||
2962507c SW |
1227 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
1228 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
1229 | else | |
1230 | want = CEPH_CAP_FILE_CACHE; | |
3738daa6 | 1231 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page); |
124e68e7 | 1232 | if (ret < 0) |
8eb4efb0 | 1233 | return ret; |
124e68e7 | 1234 | |
2962507c | 1235 | if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 || |
2ba48ce5 | 1236 | (iocb->ki_flags & IOCB_DIRECT) || |
8eb4efb0 | 1237 | (fi->flags & CEPH_F_SYNC)) { |
8eb4efb0 | 1238 | |
1239 | dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n", | |
1240 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, | |
1241 | ceph_cap_string(got)); | |
1242 | ||
83701246 | 1243 | if (ci->i_inline_version == CEPH_INLINE_NONE) { |
c8fe9b17 YZ |
1244 | if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) { |
1245 | ret = ceph_direct_read_write(iocb, to, | |
1246 | NULL, NULL); | |
1247 | if (ret >= 0 && ret < len) | |
1248 | retry_op = CHECK_EOF; | |
1249 | } else { | |
1250 | ret = ceph_sync_read(iocb, to, &retry_op); | |
1251 | } | |
83701246 YZ |
1252 | } else { |
1253 | retry_op = READ_INLINE; | |
1254 | } | |
8eb4efb0 | 1255 | } else { |
8eb4efb0 | 1256 | dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n", |
3644424d | 1257 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, |
8eb4efb0 | 1258 | ceph_cap_string(got)); |
2b1ac852 | 1259 | current->journal_info = filp; |
3644424d | 1260 | ret = generic_file_read_iter(iocb, to); |
2b1ac852 | 1261 | current->journal_info = NULL; |
8eb4efb0 | 1262 | } |
124e68e7 SW |
1263 | dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", |
1264 | inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); | |
3738daa6 | 1265 | if (pinned_page) { |
09cbfeaf | 1266 | put_page(pinned_page); |
3738daa6 YZ |
1267 | pinned_page = NULL; |
1268 | } | |
124e68e7 | 1269 | ceph_put_cap_refs(ci, got); |
c8fe9b17 | 1270 | if (retry_op > HAVE_RETRIED && ret >= 0) { |
83701246 YZ |
1271 | int statret; |
1272 | struct page *page = NULL; | |
1273 | loff_t i_size; | |
1274 | if (retry_op == READ_INLINE) { | |
687265e5 | 1275 | page = __page_cache_alloc(GFP_KERNEL); |
83701246 YZ |
1276 | if (!page) |
1277 | return -ENOMEM; | |
1278 | } | |
6a026589 | 1279 | |
83701246 YZ |
1280 | statret = __ceph_do_getattr(inode, page, |
1281 | CEPH_STAT_CAP_INLINE_DATA, !!page); | |
1282 | if (statret < 0) { | |
0d7718f6 NB |
1283 | if (page) |
1284 | __free_page(page); | |
83701246 YZ |
1285 | if (statret == -ENODATA) { |
1286 | BUG_ON(retry_op != READ_INLINE); | |
1287 | goto again; | |
1288 | } | |
1289 | return statret; | |
1290 | } | |
6a026589 | 1291 | |
83701246 YZ |
1292 | i_size = i_size_read(inode); |
1293 | if (retry_op == READ_INLINE) { | |
fcc02d2a YZ |
1294 | BUG_ON(ret > 0 || read > 0); |
1295 | if (iocb->ki_pos < i_size && | |
09cbfeaf | 1296 | iocb->ki_pos < PAGE_SIZE) { |
83701246 YZ |
1297 | loff_t end = min_t(loff_t, i_size, |
1298 | iocb->ki_pos + len); | |
09cbfeaf | 1299 | end = min_t(loff_t, end, PAGE_SIZE); |
83701246 YZ |
1300 | if (statret < end) |
1301 | zero_user_segment(page, statret, end); | |
1302 | ret = copy_page_to_iter(page, | |
1303 | iocb->ki_pos & ~PAGE_MASK, | |
1304 | end - iocb->ki_pos, to); | |
1305 | iocb->ki_pos += ret; | |
fcc02d2a YZ |
1306 | read += ret; |
1307 | } | |
1308 | if (iocb->ki_pos < i_size && read < len) { | |
1309 | size_t zlen = min_t(size_t, len - read, | |
1310 | i_size - iocb->ki_pos); | |
1311 | ret = iov_iter_zero(zlen, to); | |
1312 | iocb->ki_pos += ret; | |
1313 | read += ret; | |
83701246 YZ |
1314 | } |
1315 | __free_pages(page, 0); | |
fcc02d2a | 1316 | return read; |
83701246 | 1317 | } |
6a026589 SW |
1318 | |
1319 | /* hit EOF or hole? */ | |
83701246 | 1320 | if (retry_op == CHECK_EOF && iocb->ki_pos < i_size && |
fcc02d2a | 1321 | ret < len) { |
8eb4efb0 | 1322 | dout("sync_read hit hole, ppos %lld < size %lld" |
99c88e69 | 1323 | ", reading more\n", iocb->ki_pos, i_size); |
8eb4efb0 | 1324 | |
6a026589 | 1325 | read += ret; |
6a026589 | 1326 | len -= ret; |
c8fe9b17 | 1327 | retry_op = HAVE_RETRIED; |
6a026589 SW |
1328 | goto again; |
1329 | } | |
1330 | } | |
8eb4efb0 | 1331 | |
6a026589 SW |
1332 | if (ret >= 0) |
1333 | ret += read; | |
1334 | ||
124e68e7 SW |
1335 | return ret; |
1336 | } | |
1337 | ||
1338 | /* | |
1339 | * Take cap references to avoid releasing caps to MDS mid-write. | |
1340 | * | |
1341 | * If we are synchronous, and write with an old snap context, the OSD | |
1342 | * may return EOLDSNAPC. In that case, retry the write.. _after_ | |
1343 | * dropping our cap refs and allowing the pending snap to logically | |
1344 | * complete _before_ this write occurs. | |
1345 | * | |
1346 | * If we are near ENOSPC, write synchronously. | |
1347 | */ | |
4908b822 | 1348 | static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from) |
124e68e7 SW |
1349 | { |
1350 | struct file *file = iocb->ki_filp; | |
33caad32 | 1351 | struct ceph_file_info *fi = file->private_data; |
496ad9aa | 1352 | struct inode *inode = file_inode(file); |
124e68e7 | 1353 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 YS |
1354 | struct ceph_osd_client *osdc = |
1355 | &ceph_sb_to_client(inode->i_sb)->client->osdc; | |
f66fd9f0 | 1356 | struct ceph_cap_flush *prealloc_cf; |
3309dd04 | 1357 | ssize_t count, written = 0; |
03d254ed | 1358 | int err, want, got; |
3309dd04 | 1359 | loff_t pos; |
124e68e7 SW |
1360 | |
1361 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
1362 | return -EROFS; | |
1363 | ||
f66fd9f0 YZ |
1364 | prealloc_cf = ceph_alloc_cap_flush(); |
1365 | if (!prealloc_cf) | |
1366 | return -ENOMEM; | |
1367 | ||
5955102c | 1368 | inode_lock(inode); |
03d254ed | 1369 | |
03d254ed | 1370 | /* We can write back this queue in page reclaim */ |
de1414a6 | 1371 | current->backing_dev_info = inode_to_bdi(inode); |
03d254ed | 1372 | |
55b0b31c YZ |
1373 | if (iocb->ki_flags & IOCB_APPEND) { |
1374 | err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); | |
1375 | if (err < 0) | |
1376 | goto out; | |
1377 | } | |
1378 | ||
3309dd04 AV |
1379 | err = generic_write_checks(iocb, from); |
1380 | if (err <= 0) | |
03d254ed YZ |
1381 | goto out; |
1382 | ||
3309dd04 AV |
1383 | pos = iocb->ki_pos; |
1384 | count = iov_iter_count(from); | |
5fa8e0a1 | 1385 | err = file_remove_privs(file); |
03d254ed YZ |
1386 | if (err) |
1387 | goto out; | |
1388 | ||
1389 | err = file_update_time(file); | |
1390 | if (err) | |
1391 | goto out; | |
1392 | ||
28127bdd YZ |
1393 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
1394 | err = ceph_uninline_data(file, NULL); | |
1395 | if (err < 0) | |
1396 | goto out; | |
1397 | } | |
1398 | ||
124e68e7 | 1399 | retry_snap: |
b7ec35b3 | 1400 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL)) { |
03d254ed | 1401 | err = -ENOSPC; |
6070e0c1 YZ |
1402 | goto out; |
1403 | } | |
03d254ed | 1404 | |
ac7f29bf | 1405 | dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n", |
99c88e69 | 1406 | inode, ceph_vinop(inode), pos, count, i_size_read(inode)); |
7971bd92 SW |
1407 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
1408 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1409 | else | |
1410 | want = CEPH_CAP_FILE_BUFFER; | |
03d254ed | 1411 | got = 0; |
3738daa6 YZ |
1412 | err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count, |
1413 | &got, NULL); | |
03d254ed | 1414 | if (err < 0) |
37505d57 | 1415 | goto out; |
124e68e7 | 1416 | |
ac7f29bf | 1417 | dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n", |
03d254ed | 1418 | inode, ceph_vinop(inode), pos, count, ceph_cap_string(got)); |
7971bd92 SW |
1419 | |
1420 | if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || | |
2ba48ce5 | 1421 | (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC)) { |
5dda377c | 1422 | struct ceph_snap_context *snapc; |
4908b822 | 1423 | struct iov_iter data; |
5955102c | 1424 | inode_unlock(inode); |
5dda377c YZ |
1425 | |
1426 | spin_lock(&ci->i_ceph_lock); | |
1427 | if (__ceph_have_pending_cap_snap(ci)) { | |
1428 | struct ceph_cap_snap *capsnap = | |
1429 | list_last_entry(&ci->i_cap_snaps, | |
1430 | struct ceph_cap_snap, | |
1431 | ci_item); | |
1432 | snapc = ceph_get_snap_context(capsnap->context); | |
1433 | } else { | |
1434 | BUG_ON(!ci->i_head_snapc); | |
1435 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
1436 | } | |
1437 | spin_unlock(&ci->i_ceph_lock); | |
1438 | ||
4908b822 AV |
1439 | /* we might need to revert back to that point */ |
1440 | data = *from; | |
2ba48ce5 | 1441 | if (iocb->ki_flags & IOCB_DIRECT) |
c8fe9b17 YZ |
1442 | written = ceph_direct_read_write(iocb, &data, snapc, |
1443 | &prealloc_cf); | |
e8344e66 | 1444 | else |
5dda377c | 1445 | written = ceph_sync_write(iocb, &data, pos, snapc); |
0e5dd45c | 1446 | if (written == -EOLDSNAPC) { |
1447 | dout("aio_write %p %llx.%llx %llu~%u" | |
1448 | "got EOLDSNAPC, retrying\n", | |
1449 | inode, ceph_vinop(inode), | |
4908b822 | 1450 | pos, (unsigned)count); |
5955102c | 1451 | inode_lock(inode); |
0e5dd45c | 1452 | goto retry_snap; |
1453 | } | |
4908b822 AV |
1454 | if (written > 0) |
1455 | iov_iter_advance(from, written); | |
5dda377c | 1456 | ceph_put_snap_context(snapc); |
7971bd92 | 1457 | } else { |
b0d7c223 YZ |
1458 | /* |
1459 | * No need to acquire the i_truncate_mutex. Because | |
1460 | * the MDS revokes Fwb caps before sending truncate | |
1461 | * message to us. We can't get Fwb cap while there | |
1462 | * are pending vmtruncate. So write and vmtruncate | |
1463 | * can not run at the same time | |
1464 | */ | |
4908b822 | 1465 | written = generic_perform_write(file, from, pos); |
aec605f4 AV |
1466 | if (likely(written >= 0)) |
1467 | iocb->ki_pos = pos + written; | |
5955102c | 1468 | inode_unlock(inode); |
7971bd92 | 1469 | } |
d8de9ab6 | 1470 | |
03d254ed | 1471 | if (written >= 0) { |
fca65b4a | 1472 | int dirty; |
be655596 | 1473 | spin_lock(&ci->i_ceph_lock); |
28127bdd | 1474 | ci->i_inline_version = CEPH_INLINE_NONE; |
f66fd9f0 YZ |
1475 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
1476 | &prealloc_cf); | |
be655596 | 1477 | spin_unlock(&ci->i_ceph_lock); |
fca65b4a SW |
1478 | if (dirty) |
1479 | __mark_inode_dirty(inode, dirty); | |
124e68e7 | 1480 | } |
7971bd92 | 1481 | |
124e68e7 | 1482 | dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n", |
4908b822 | 1483 | inode, ceph_vinop(inode), pos, (unsigned)count, |
7971bd92 | 1484 | ceph_cap_string(got)); |
124e68e7 | 1485 | ceph_put_cap_refs(ci, got); |
7971bd92 | 1486 | |
6aa657c8 | 1487 | if (written >= 0) { |
b7ec35b3 | 1488 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_NEARFULL)) |
6aa657c8 CH |
1489 | iocb->ki_flags |= IOCB_DSYNC; |
1490 | ||
1491 | written = generic_write_sync(iocb, written); | |
6070e0c1 | 1492 | } |
03d254ed | 1493 | |
2f75e9e1 SW |
1494 | goto out_unlocked; |
1495 | ||
03d254ed | 1496 | out: |
5955102c | 1497 | inode_unlock(inode); |
2f75e9e1 | 1498 | out_unlocked: |
f66fd9f0 | 1499 | ceph_free_cap_flush(prealloc_cf); |
03d254ed | 1500 | current->backing_dev_info = NULL; |
03d254ed | 1501 | return written ? written : err; |
124e68e7 SW |
1502 | } |
1503 | ||
1504 | /* | |
1505 | * llseek. be sure to verify file size on SEEK_END. | |
1506 | */ | |
965c8e59 | 1507 | static loff_t ceph_llseek(struct file *file, loff_t offset, int whence) |
124e68e7 SW |
1508 | { |
1509 | struct inode *inode = file->f_mapping->host; | |
99c88e69 | 1510 | loff_t i_size; |
955818cd | 1511 | loff_t ret; |
124e68e7 | 1512 | |
5955102c | 1513 | inode_lock(inode); |
6a82c47a | 1514 | |
965c8e59 | 1515 | if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { |
508b32d8 | 1516 | ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); |
955818cd | 1517 | if (ret < 0) |
124e68e7 | 1518 | goto out; |
06222e49 JB |
1519 | } |
1520 | ||
99c88e69 | 1521 | i_size = i_size_read(inode); |
965c8e59 | 1522 | switch (whence) { |
06222e49 | 1523 | case SEEK_END: |
99c88e69 | 1524 | offset += i_size; |
124e68e7 SW |
1525 | break; |
1526 | case SEEK_CUR: | |
1527 | /* | |
1528 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
1529 | * position-querying operation. Avoid rewriting the "same" | |
1530 | * f_pos value back to the file because a concurrent read(), | |
1531 | * write() or lseek() might have altered it | |
1532 | */ | |
1533 | if (offset == 0) { | |
955818cd | 1534 | ret = file->f_pos; |
124e68e7 SW |
1535 | goto out; |
1536 | } | |
1537 | offset += file->f_pos; | |
1538 | break; | |
06222e49 | 1539 | case SEEK_DATA: |
99c88e69 | 1540 | if (offset >= i_size) { |
06222e49 JB |
1541 | ret = -ENXIO; |
1542 | goto out; | |
1543 | } | |
1544 | break; | |
1545 | case SEEK_HOLE: | |
99c88e69 | 1546 | if (offset >= i_size) { |
06222e49 JB |
1547 | ret = -ENXIO; |
1548 | goto out; | |
1549 | } | |
99c88e69 | 1550 | offset = i_size; |
06222e49 | 1551 | break; |
124e68e7 SW |
1552 | } |
1553 | ||
955818cd | 1554 | ret = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
124e68e7 SW |
1555 | |
1556 | out: | |
5955102c | 1557 | inode_unlock(inode); |
955818cd | 1558 | return ret; |
124e68e7 SW |
1559 | } |
1560 | ||
ad7a60de LW |
1561 | static inline void ceph_zero_partial_page( |
1562 | struct inode *inode, loff_t offset, unsigned size) | |
1563 | { | |
1564 | struct page *page; | |
09cbfeaf | 1565 | pgoff_t index = offset >> PAGE_SHIFT; |
ad7a60de LW |
1566 | |
1567 | page = find_lock_page(inode->i_mapping, index); | |
1568 | if (page) { | |
1569 | wait_on_page_writeback(page); | |
09cbfeaf | 1570 | zero_user(page, offset & (PAGE_SIZE - 1), size); |
ad7a60de | 1571 | unlock_page(page); |
09cbfeaf | 1572 | put_page(page); |
ad7a60de LW |
1573 | } |
1574 | } | |
1575 | ||
1576 | static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset, | |
1577 | loff_t length) | |
1578 | { | |
09cbfeaf | 1579 | loff_t nearly = round_up(offset, PAGE_SIZE); |
ad7a60de LW |
1580 | if (offset < nearly) { |
1581 | loff_t size = nearly - offset; | |
1582 | if (length < size) | |
1583 | size = length; | |
1584 | ceph_zero_partial_page(inode, offset, size); | |
1585 | offset += size; | |
1586 | length -= size; | |
1587 | } | |
09cbfeaf KS |
1588 | if (length >= PAGE_SIZE) { |
1589 | loff_t size = round_down(length, PAGE_SIZE); | |
ad7a60de LW |
1590 | truncate_pagecache_range(inode, offset, offset + size - 1); |
1591 | offset += size; | |
1592 | length -= size; | |
1593 | } | |
1594 | if (length) | |
1595 | ceph_zero_partial_page(inode, offset, length); | |
1596 | } | |
1597 | ||
1598 | static int ceph_zero_partial_object(struct inode *inode, | |
1599 | loff_t offset, loff_t *length) | |
1600 | { | |
1601 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1602 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
1603 | struct ceph_osd_request *req; | |
1604 | int ret = 0; | |
1605 | loff_t zero = 0; | |
1606 | int op; | |
1607 | ||
1608 | if (!length) { | |
1609 | op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE; | |
1610 | length = &zero; | |
1611 | } else { | |
1612 | op = CEPH_OSD_OP_ZERO; | |
1613 | } | |
1614 | ||
1615 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1616 | ceph_vino(inode), | |
1617 | offset, length, | |
715e4cd4 | 1618 | 0, 1, op, |
ad7a60de LW |
1619 | CEPH_OSD_FLAG_WRITE | |
1620 | CEPH_OSD_FLAG_ONDISK, | |
1621 | NULL, 0, 0, false); | |
1622 | if (IS_ERR(req)) { | |
1623 | ret = PTR_ERR(req); | |
1624 | goto out; | |
1625 | } | |
1626 | ||
bb873b53 | 1627 | req->r_mtime = inode->i_mtime; |
ad7a60de LW |
1628 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
1629 | if (!ret) { | |
1630 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1631 | if (ret == -ENOENT) | |
1632 | ret = 0; | |
1633 | } | |
1634 | ceph_osdc_put_request(req); | |
1635 | ||
1636 | out: | |
1637 | return ret; | |
1638 | } | |
1639 | ||
1640 | static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length) | |
1641 | { | |
1642 | int ret = 0; | |
1643 | struct ceph_inode_info *ci = ceph_inode(inode); | |
7627151e YZ |
1644 | s32 stripe_unit = ci->i_layout.stripe_unit; |
1645 | s32 stripe_count = ci->i_layout.stripe_count; | |
1646 | s32 object_size = ci->i_layout.object_size; | |
b314a90d SW |
1647 | u64 object_set_size = object_size * stripe_count; |
1648 | u64 nearly, t; | |
1649 | ||
1650 | /* round offset up to next period boundary */ | |
1651 | nearly = offset + object_set_size - 1; | |
1652 | t = nearly; | |
1653 | nearly -= do_div(t, object_set_size); | |
ad7a60de | 1654 | |
ad7a60de LW |
1655 | while (length && offset < nearly) { |
1656 | loff_t size = length; | |
1657 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1658 | if (ret < 0) | |
1659 | return ret; | |
1660 | offset += size; | |
1661 | length -= size; | |
1662 | } | |
1663 | while (length >= object_set_size) { | |
1664 | int i; | |
1665 | loff_t pos = offset; | |
1666 | for (i = 0; i < stripe_count; ++i) { | |
1667 | ret = ceph_zero_partial_object(inode, pos, NULL); | |
1668 | if (ret < 0) | |
1669 | return ret; | |
1670 | pos += stripe_unit; | |
1671 | } | |
1672 | offset += object_set_size; | |
1673 | length -= object_set_size; | |
1674 | } | |
1675 | while (length) { | |
1676 | loff_t size = length; | |
1677 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1678 | if (ret < 0) | |
1679 | return ret; | |
1680 | offset += size; | |
1681 | length -= size; | |
1682 | } | |
1683 | return ret; | |
1684 | } | |
1685 | ||
1686 | static long ceph_fallocate(struct file *file, int mode, | |
1687 | loff_t offset, loff_t length) | |
1688 | { | |
1689 | struct ceph_file_info *fi = file->private_data; | |
aa8b60e0 | 1690 | struct inode *inode = file_inode(file); |
ad7a60de LW |
1691 | struct ceph_inode_info *ci = ceph_inode(inode); |
1692 | struct ceph_osd_client *osdc = | |
1693 | &ceph_inode_to_client(inode)->client->osdc; | |
f66fd9f0 | 1694 | struct ceph_cap_flush *prealloc_cf; |
ad7a60de LW |
1695 | int want, got = 0; |
1696 | int dirty; | |
1697 | int ret = 0; | |
1698 | loff_t endoff = 0; | |
1699 | loff_t size; | |
1700 | ||
494d77bf YZ |
1701 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
1702 | return -EOPNOTSUPP; | |
1703 | ||
ad7a60de LW |
1704 | if (!S_ISREG(inode->i_mode)) |
1705 | return -EOPNOTSUPP; | |
1706 | ||
f66fd9f0 YZ |
1707 | prealloc_cf = ceph_alloc_cap_flush(); |
1708 | if (!prealloc_cf) | |
1709 | return -ENOMEM; | |
1710 | ||
5955102c | 1711 | inode_lock(inode); |
ad7a60de LW |
1712 | |
1713 | if (ceph_snap(inode) != CEPH_NOSNAP) { | |
1714 | ret = -EROFS; | |
1715 | goto unlock; | |
1716 | } | |
1717 | ||
b7ec35b3 ID |
1718 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) && |
1719 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
ad7a60de LW |
1720 | ret = -ENOSPC; |
1721 | goto unlock; | |
1722 | } | |
1723 | ||
28127bdd YZ |
1724 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
1725 | ret = ceph_uninline_data(file, NULL); | |
1726 | if (ret < 0) | |
1727 | goto unlock; | |
1728 | } | |
1729 | ||
ad7a60de LW |
1730 | size = i_size_read(inode); |
1731 | if (!(mode & FALLOC_FL_KEEP_SIZE)) | |
1732 | endoff = offset + length; | |
1733 | ||
1734 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1735 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1736 | else | |
1737 | want = CEPH_CAP_FILE_BUFFER; | |
1738 | ||
3738daa6 | 1739 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL); |
ad7a60de LW |
1740 | if (ret < 0) |
1741 | goto unlock; | |
1742 | ||
1743 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1744 | if (offset < size) | |
1745 | ceph_zero_pagecache_range(inode, offset, length); | |
1746 | ret = ceph_zero_objects(inode, offset, length); | |
1747 | } else if (endoff > size) { | |
1748 | truncate_pagecache_range(inode, size, -1); | |
1749 | if (ceph_inode_set_size(inode, endoff)) | |
1750 | ceph_check_caps(ceph_inode(inode), | |
1751 | CHECK_CAPS_AUTHONLY, NULL); | |
1752 | } | |
1753 | ||
1754 | if (!ret) { | |
1755 | spin_lock(&ci->i_ceph_lock); | |
28127bdd | 1756 | ci->i_inline_version = CEPH_INLINE_NONE; |
f66fd9f0 YZ |
1757 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
1758 | &prealloc_cf); | |
ad7a60de LW |
1759 | spin_unlock(&ci->i_ceph_lock); |
1760 | if (dirty) | |
1761 | __mark_inode_dirty(inode, dirty); | |
1762 | } | |
1763 | ||
1764 | ceph_put_cap_refs(ci, got); | |
1765 | unlock: | |
5955102c | 1766 | inode_unlock(inode); |
f66fd9f0 | 1767 | ceph_free_cap_flush(prealloc_cf); |
ad7a60de LW |
1768 | return ret; |
1769 | } | |
1770 | ||
124e68e7 SW |
1771 | const struct file_operations ceph_file_fops = { |
1772 | .open = ceph_open, | |
1773 | .release = ceph_release, | |
1774 | .llseek = ceph_llseek, | |
3644424d | 1775 | .read_iter = ceph_read_iter, |
4908b822 | 1776 | .write_iter = ceph_write_iter, |
124e68e7 SW |
1777 | .mmap = ceph_mmap, |
1778 | .fsync = ceph_fsync, | |
40819f6f GF |
1779 | .lock = ceph_lock, |
1780 | .flock = ceph_flock, | |
7ce469a5 | 1781 | .splice_read = generic_file_splice_read, |
3551dd79 | 1782 | .splice_write = iter_file_splice_write, |
124e68e7 SW |
1783 | .unlocked_ioctl = ceph_ioctl, |
1784 | .compat_ioctl = ceph_ioctl, | |
ad7a60de | 1785 | .fallocate = ceph_fallocate, |
124e68e7 SW |
1786 | }; |
1787 |