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; | |
99ccbd22 MT |
140 | struct ceph_inode_info *ci = ceph_inode(inode); |
141 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); | |
142 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
143 | |
144 | switch (inode->i_mode & S_IFMT) { | |
145 | case S_IFREG: | |
99ccbd22 MT |
146 | /* First file open request creates the cookie, we want to keep |
147 | * this cookie around for the filetime of the inode as not to | |
148 | * have to worry about fscache register / revoke / operation | |
149 | * races. | |
150 | * | |
151 | * Also, if we know the operation is going to invalidate data | |
152 | * (non readonly) just nuke the cache right away. | |
153 | */ | |
154 | ceph_fscache_register_inode_cookie(mdsc->fsc, ci); | |
155 | if ((fmode & CEPH_FILE_MODE_WR)) | |
156 | ceph_fscache_invalidate(inode); | |
124e68e7 SW |
157 | case S_IFDIR: |
158 | dout("init_file %p %p 0%o (regular)\n", inode, file, | |
159 | inode->i_mode); | |
99ec2697 | 160 | cf = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL); |
124e68e7 SW |
161 | if (cf == NULL) { |
162 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
163 | return -ENOMEM; | |
164 | } | |
165 | cf->fmode = fmode; | |
166 | cf->next_offset = 2; | |
fdd4e158 | 167 | cf->readdir_cache_idx = -1; |
124e68e7 SW |
168 | file->private_data = cf; |
169 | BUG_ON(inode->i_fop->release != ceph_release); | |
170 | break; | |
171 | ||
172 | case S_IFLNK: | |
173 | dout("init_file %p %p 0%o (symlink)\n", inode, file, | |
174 | inode->i_mode); | |
175 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
176 | break; | |
177 | ||
178 | default: | |
179 | dout("init_file %p %p 0%o (special)\n", inode, file, | |
180 | inode->i_mode); | |
181 | /* | |
182 | * we need to drop the open ref now, since we don't | |
183 | * have .release set to ceph_release. | |
184 | */ | |
185 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
186 | BUG_ON(inode->i_fop->release == ceph_release); | |
187 | ||
188 | /* call the proper open fop */ | |
189 | ret = inode->i_fop->open(inode, file); | |
190 | } | |
191 | return ret; | |
192 | } | |
193 | ||
77310320 YZ |
194 | /* |
195 | * try renew caps after session gets killed. | |
196 | */ | |
197 | int ceph_renew_caps(struct inode *inode) | |
198 | { | |
199 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; | |
200 | struct ceph_inode_info *ci = ceph_inode(inode); | |
201 | struct ceph_mds_request *req; | |
202 | int err, flags, wanted; | |
203 | ||
204 | spin_lock(&ci->i_ceph_lock); | |
205 | wanted = __ceph_caps_file_wanted(ci); | |
206 | if (__ceph_is_any_real_caps(ci) && | |
207 | (!(wanted & CEPH_CAP_ANY_WR) == 0 || ci->i_auth_cap)) { | |
208 | int issued = __ceph_caps_issued(ci, NULL); | |
209 | spin_unlock(&ci->i_ceph_lock); | |
210 | dout("renew caps %p want %s issued %s updating mds_wanted\n", | |
211 | inode, ceph_cap_string(wanted), ceph_cap_string(issued)); | |
212 | ceph_check_caps(ci, 0, NULL); | |
213 | return 0; | |
214 | } | |
215 | spin_unlock(&ci->i_ceph_lock); | |
216 | ||
217 | flags = 0; | |
218 | if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR)) | |
219 | flags = O_RDWR; | |
220 | else if (wanted & CEPH_CAP_FILE_RD) | |
221 | flags = O_RDONLY; | |
222 | else if (wanted & CEPH_CAP_FILE_WR) | |
223 | flags = O_WRONLY; | |
224 | #ifdef O_LAZY | |
225 | if (wanted & CEPH_CAP_FILE_LAZYIO) | |
226 | flags |= O_LAZY; | |
227 | #endif | |
228 | ||
229 | req = prepare_open_request(inode->i_sb, flags, 0); | |
230 | if (IS_ERR(req)) { | |
231 | err = PTR_ERR(req); | |
232 | goto out; | |
233 | } | |
234 | ||
235 | req->r_inode = inode; | |
236 | ihold(inode); | |
237 | req->r_num_caps = 1; | |
238 | req->r_fmode = -1; | |
239 | ||
240 | err = ceph_mdsc_do_request(mdsc, NULL, req); | |
241 | ceph_mdsc_put_request(req); | |
242 | out: | |
243 | dout("renew caps %p open result=%d\n", inode, err); | |
244 | return err < 0 ? err : 0; | |
245 | } | |
246 | ||
124e68e7 | 247 | /* |
124e68e7 SW |
248 | * If we already have the requisite capabilities, we can satisfy |
249 | * the open request locally (no need to request new caps from the | |
250 | * MDS). We do, however, need to inform the MDS (asynchronously) | |
251 | * if our wanted caps set expands. | |
252 | */ | |
253 | int ceph_open(struct inode *inode, struct file *file) | |
254 | { | |
255 | struct ceph_inode_info *ci = ceph_inode(inode); | |
3d14c5d2 YS |
256 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
257 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
258 | struct ceph_mds_request *req; |
259 | struct ceph_file_info *cf = file->private_data; | |
124e68e7 SW |
260 | int err; |
261 | int flags, fmode, wanted; | |
262 | ||
263 | if (cf) { | |
264 | dout("open file %p is already opened\n", file); | |
265 | return 0; | |
266 | } | |
267 | ||
268 | /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */ | |
269 | flags = file->f_flags & ~(O_CREAT|O_EXCL); | |
270 | if (S_ISDIR(inode->i_mode)) | |
271 | flags = O_DIRECTORY; /* mds likes to know */ | |
272 | ||
273 | dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode, | |
274 | ceph_vinop(inode), file, flags, file->f_flags); | |
275 | fmode = ceph_flags_to_mode(flags); | |
276 | wanted = ceph_caps_for_mode(fmode); | |
277 | ||
278 | /* snapped files are read-only */ | |
279 | if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE)) | |
280 | return -EROFS; | |
281 | ||
282 | /* trivially open snapdir */ | |
283 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
be655596 | 284 | spin_lock(&ci->i_ceph_lock); |
124e68e7 | 285 | __ceph_get_fmode(ci, fmode); |
be655596 | 286 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
287 | return ceph_init_file(inode, file, fmode); |
288 | } | |
289 | ||
290 | /* | |
7421ab80 SW |
291 | * No need to block if we have caps on the auth MDS (for |
292 | * write) or any MDS (for read). Update wanted set | |
124e68e7 SW |
293 | * asynchronously. |
294 | */ | |
be655596 | 295 | spin_lock(&ci->i_ceph_lock); |
7421ab80 SW |
296 | if (__ceph_is_any_real_caps(ci) && |
297 | (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) { | |
124e68e7 SW |
298 | int mds_wanted = __ceph_caps_mds_wanted(ci); |
299 | int issued = __ceph_caps_issued(ci, NULL); | |
300 | ||
301 | dout("open %p fmode %d want %s issued %s using existing\n", | |
302 | inode, fmode, ceph_cap_string(wanted), | |
303 | ceph_cap_string(issued)); | |
304 | __ceph_get_fmode(ci, fmode); | |
be655596 | 305 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
306 | |
307 | /* adjust wanted? */ | |
308 | if ((issued & wanted) != wanted && | |
309 | (mds_wanted & wanted) != wanted && | |
310 | ceph_snap(inode) != CEPH_SNAPDIR) | |
311 | ceph_check_caps(ci, 0, NULL); | |
312 | ||
313 | return ceph_init_file(inode, file, fmode); | |
314 | } else if (ceph_snap(inode) != CEPH_NOSNAP && | |
315 | (ci->i_snap_caps & wanted) == wanted) { | |
316 | __ceph_get_fmode(ci, fmode); | |
be655596 | 317 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
318 | return ceph_init_file(inode, file, fmode); |
319 | } | |
99ccbd22 | 320 | |
be655596 | 321 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
322 | |
323 | dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted)); | |
324 | req = prepare_open_request(inode->i_sb, flags, 0); | |
325 | if (IS_ERR(req)) { | |
326 | err = PTR_ERR(req); | |
327 | goto out; | |
328 | } | |
70b666c3 SW |
329 | req->r_inode = inode; |
330 | ihold(inode); | |
99ccbd22 | 331 | |
124e68e7 | 332 | req->r_num_caps = 1; |
e36d571d | 333 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
124e68e7 SW |
334 | if (!err) |
335 | err = ceph_init_file(inode, file, req->r_fmode); | |
336 | ceph_mdsc_put_request(req); | |
337 | dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode)); | |
338 | out: | |
339 | return err; | |
340 | } | |
341 | ||
342 | ||
343 | /* | |
5ef50c3b SW |
344 | * Do a lookup + open with a single request. If we get a non-existent |
345 | * file or symlink, return 1 so the VFS can retry. | |
124e68e7 | 346 | */ |
5ef50c3b | 347 | int ceph_atomic_open(struct inode *dir, struct dentry *dentry, |
30d90494 | 348 | struct file *file, unsigned flags, umode_t mode, |
d9585277 | 349 | int *opened) |
124e68e7 | 350 | { |
3d14c5d2 YS |
351 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
352 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 | 353 | struct ceph_mds_request *req; |
5ef50c3b | 354 | struct dentry *dn; |
b1ee94aa | 355 | struct ceph_acls_info acls = {}; |
315f2408 | 356 | int mask; |
124e68e7 | 357 | int err; |
124e68e7 | 358 | |
a455589f AV |
359 | dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n", |
360 | dir, dentry, dentry, | |
5ef50c3b SW |
361 | d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode); |
362 | ||
363 | if (dentry->d_name.len > NAME_MAX) | |
364 | return -ENAMETOOLONG; | |
365 | ||
366 | err = ceph_init_dentry(dentry); | |
367 | if (err < 0) | |
368 | return err; | |
124e68e7 | 369 | |
b1ee94aa YZ |
370 | if (flags & O_CREAT) { |
371 | err = ceph_pre_init_acls(dir, &mode, &acls); | |
372 | if (err < 0) | |
373 | return err; | |
374 | } | |
375 | ||
124e68e7 SW |
376 | /* do the open */ |
377 | req = prepare_open_request(dir->i_sb, flags, mode); | |
b1ee94aa YZ |
378 | if (IS_ERR(req)) { |
379 | err = PTR_ERR(req); | |
380 | goto out_acl; | |
381 | } | |
124e68e7 SW |
382 | req->r_dentry = dget(dentry); |
383 | req->r_num_caps = 2; | |
384 | if (flags & O_CREAT) { | |
385 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
386 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
b1ee94aa YZ |
387 | if (acls.pagelist) { |
388 | req->r_pagelist = acls.pagelist; | |
389 | acls.pagelist = NULL; | |
390 | } | |
124e68e7 | 391 | } |
315f2408 YZ |
392 | |
393 | mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED; | |
394 | if (ceph_security_xattr_wanted(dir)) | |
395 | mask |= CEPH_CAP_XATTR_SHARED; | |
396 | req->r_args.open.mask = cpu_to_le32(mask); | |
397 | ||
124e68e7 | 398 | req->r_locked_dir = dir; /* caller holds dir->i_mutex */ |
acda7657 SW |
399 | err = ceph_mdsc_do_request(mdsc, |
400 | (flags & (O_CREAT|O_TRUNC)) ? dir : NULL, | |
401 | req); | |
bf91c315 | 402 | err = ceph_handle_snapdir(req, dentry, err); |
79aec984 | 403 | if (err) |
b1ee94aa | 404 | goto out_req; |
79aec984 | 405 | |
a43137f7 | 406 | if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry) |
124e68e7 | 407 | err = ceph_handle_notrace_create(dir, dentry); |
2d83bde9 | 408 | |
5ef50c3b SW |
409 | if (d_unhashed(dentry)) { |
410 | dn = ceph_finish_lookup(req, dentry, err); | |
411 | if (IS_ERR(dn)) | |
412 | err = PTR_ERR(dn); | |
413 | } else { | |
414 | /* we were given a hashed negative dentry */ | |
415 | dn = NULL; | |
416 | } | |
417 | if (err) | |
b1ee94aa | 418 | goto out_req; |
2b0143b5 | 419 | if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) { |
5ef50c3b SW |
420 | /* make vfs retry on splice, ENOENT, or symlink */ |
421 | dout("atomic_open finish_no_open on dn %p\n", dn); | |
422 | err = finish_no_open(file, dn); | |
423 | } else { | |
424 | dout("atomic_open finish_open on dn %p\n", dn); | |
6e8575fa | 425 | if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) { |
2b0143b5 | 426 | ceph_init_inode_acls(d_inode(dentry), &acls); |
6e8575fa SL |
427 | *opened |= FILE_CREATED; |
428 | } | |
5ef50c3b SW |
429 | err = finish_open(file, dentry, ceph_open, opened); |
430 | } | |
b1ee94aa | 431 | out_req: |
ab866549 YZ |
432 | if (!req->r_err && req->r_target_inode) |
433 | ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode); | |
5ef50c3b | 434 | ceph_mdsc_put_request(req); |
b1ee94aa YZ |
435 | out_acl: |
436 | ceph_release_acls_info(&acls); | |
5ef50c3b | 437 | dout("atomic_open result=%d\n", err); |
d9585277 | 438 | return err; |
124e68e7 SW |
439 | } |
440 | ||
441 | int ceph_release(struct inode *inode, struct file *file) | |
442 | { | |
443 | struct ceph_inode_info *ci = ceph_inode(inode); | |
444 | struct ceph_file_info *cf = file->private_data; | |
445 | ||
446 | dout("release inode %p file %p\n", inode, file); | |
447 | ceph_put_fmode(ci, cf->fmode); | |
448 | if (cf->last_readdir) | |
449 | ceph_mdsc_put_request(cf->last_readdir); | |
450 | kfree(cf->last_name); | |
451 | kfree(cf->dir_info); | |
124e68e7 | 452 | kmem_cache_free(ceph_file_cachep, cf); |
195d3ce2 SW |
453 | |
454 | /* wake up anyone waiting for caps on this inode */ | |
03066f23 | 455 | wake_up_all(&ci->i_cap_wq); |
124e68e7 SW |
456 | return 0; |
457 | } | |
458 | ||
83701246 | 459 | enum { |
c8fe9b17 YZ |
460 | HAVE_RETRIED = 1, |
461 | CHECK_EOF = 2, | |
462 | READ_INLINE = 3, | |
83701246 YZ |
463 | }; |
464 | ||
124e68e7 SW |
465 | /* |
466 | * Read a range of bytes striped over one or more objects. Iterate over | |
467 | * objects we stripe over. (That's not atomic, but good enough for now.) | |
468 | * | |
469 | * If we get a short result from the OSD, check against i_size; we need to | |
470 | * only return a short read to the caller if we hit EOF. | |
471 | */ | |
472 | static int striped_read(struct inode *inode, | |
473 | u64 off, u64 len, | |
6a026589 | 474 | struct page **pages, int num_pages, |
c8fe9b17 | 475 | int *checkeof) |
124e68e7 | 476 | { |
3d14c5d2 | 477 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
124e68e7 | 478 | struct ceph_inode_info *ci = ceph_inode(inode); |
688bac46 | 479 | u64 pos, this_len, left; |
99c88e69 | 480 | loff_t i_size; |
c8fe9b17 YZ |
481 | int page_align, pages_left; |
482 | int read, ret; | |
124e68e7 | 483 | struct page **page_pos; |
124e68e7 SW |
484 | bool hit_stripe, was_short; |
485 | ||
486 | /* | |
487 | * we may need to do multiple reads. not atomic, unfortunately. | |
488 | */ | |
489 | pos = off; | |
490 | left = len; | |
491 | page_pos = pages; | |
492 | pages_left = num_pages; | |
493 | read = 0; | |
494 | ||
495 | more: | |
c8fe9b17 | 496 | page_align = pos & ~PAGE_MASK; |
124e68e7 | 497 | this_len = left; |
3d14c5d2 | 498 | ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), |
124e68e7 SW |
499 | &ci->i_layout, pos, &this_len, |
500 | ci->i_truncate_seq, | |
501 | ci->i_truncate_size, | |
b7495fc2 | 502 | page_pos, pages_left, page_align); |
124e68e7 SW |
503 | if (ret == -ENOENT) |
504 | ret = 0; | |
0e98728f SW |
505 | hit_stripe = this_len < left; |
506 | was_short = ret >= 0 && ret < this_len; | |
688bac46 | 507 | dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read, |
124e68e7 SW |
508 | ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : ""); |
509 | ||
99c88e69 | 510 | i_size = i_size_read(inode); |
02ae66d8 | 511 | if (ret >= 0) { |
512 | int didpages; | |
99c88e69 YZ |
513 | if (was_short && (pos + ret < i_size)) { |
514 | int zlen = min(this_len - ret, i_size - pos - ret); | |
c8fe9b17 | 515 | int zoff = (off & ~PAGE_MASK) + read + ret; |
02ae66d8 | 516 | dout(" zero gap %llu to %llu\n", |
1487a688 YZ |
517 | pos + ret, pos + ret + zlen); |
518 | ceph_zero_page_vector_range(zoff, zlen, pages); | |
519 | ret += zlen; | |
124e68e7 | 520 | } |
02ae66d8 | 521 | |
09cbfeaf | 522 | didpages = (page_align + ret) >> PAGE_SHIFT; |
124e68e7 SW |
523 | pos += ret; |
524 | read = pos - off; | |
525 | left -= ret; | |
526 | page_pos += didpages; | |
527 | pages_left -= didpages; | |
528 | ||
02ae66d8 | 529 | /* hit stripe and need continue*/ |
99c88e69 | 530 | if (left && hit_stripe && pos < i_size) |
124e68e7 SW |
531 | goto more; |
532 | } | |
533 | ||
ee7289bf | 534 | if (read > 0) { |
02ae66d8 | 535 | ret = read; |
c3cd6283 | 536 | /* did we bounce off eof? */ |
99c88e69 | 537 | if (pos + left > i_size) |
83701246 | 538 | *checkeof = CHECK_EOF; |
124e68e7 SW |
539 | } |
540 | ||
124e68e7 SW |
541 | dout("striped_read returns %d\n", ret); |
542 | return ret; | |
543 | } | |
544 | ||
545 | /* | |
546 | * Completely synchronous read and write methods. Direct from __user | |
547 | * buffer to osd, or directly to user pages (if O_DIRECT). | |
548 | * | |
549 | * If the read spans object boundary, just do multiple reads. | |
550 | */ | |
8eb4efb0 | 551 | static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i, |
552 | int *checkeof) | |
124e68e7 | 553 | { |
8eb4efb0 | 554 | struct file *file = iocb->ki_filp; |
496ad9aa | 555 | struct inode *inode = file_inode(file); |
124e68e7 | 556 | struct page **pages; |
8eb4efb0 | 557 | u64 off = iocb->ki_pos; |
ab226e21 | 558 | int num_pages, ret; |
2b777c9d | 559 | size_t len = iov_iter_count(i); |
124e68e7 | 560 | |
8eb4efb0 | 561 | dout("sync_read on file %p %llu~%u %s\n", file, off, |
562 | (unsigned)len, | |
124e68e7 | 563 | (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); |
d0d0db22 YZ |
564 | |
565 | if (!len) | |
566 | return 0; | |
e98b6fed SW |
567 | /* |
568 | * flush any page cache pages in this range. this | |
569 | * will make concurrent normal and sync io slow, | |
570 | * but it will at least behave sensibly when they are | |
571 | * in sequence. | |
572 | */ | |
8eb4efb0 | 573 | ret = filemap_write_and_wait_range(inode->i_mapping, off, |
574 | off + len); | |
29065a51 | 575 | if (ret < 0) |
8eb4efb0 | 576 | return ret; |
29065a51 | 577 | |
c8fe9b17 YZ |
578 | num_pages = calc_pages_for(off, len); |
579 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); | |
580 | if (IS_ERR(pages)) | |
581 | return PTR_ERR(pages); | |
582 | ret = striped_read(inode, off, len, pages, | |
583 | num_pages, checkeof); | |
584 | if (ret > 0) { | |
585 | int l, k = 0; | |
586 | size_t left = ret; | |
587 | ||
588 | while (left) { | |
589 | size_t page_off = off & ~PAGE_MASK; | |
590 | size_t copy = min_t(size_t, left, | |
591 | PAGE_SIZE - page_off); | |
592 | l = copy_page_to_iter(pages[k++], page_off, copy, i); | |
593 | off += l; | |
594 | left -= l; | |
595 | if (l < copy) | |
8eb4efb0 | 596 | break; |
597 | } | |
8eb4efb0 | 598 | } |
c8fe9b17 | 599 | ceph_release_page_vector(pages, num_pages); |
124e68e7 | 600 | |
8eb4efb0 | 601 | if (off > iocb->ki_pos) { |
602 | ret = off - iocb->ki_pos; | |
603 | iocb->ki_pos = off; | |
604 | } | |
124e68e7 | 605 | |
124e68e7 SW |
606 | dout("sync_read result %d\n", ret); |
607 | return ret; | |
608 | } | |
609 | ||
c8fe9b17 YZ |
610 | struct ceph_aio_request { |
611 | struct kiocb *iocb; | |
612 | size_t total_len; | |
613 | int write; | |
614 | int error; | |
615 | struct list_head osd_reqs; | |
616 | unsigned num_reqs; | |
617 | atomic_t pending_reqs; | |
5be0389d | 618 | struct timespec mtime; |
c8fe9b17 YZ |
619 | struct ceph_cap_flush *prealloc_cf; |
620 | }; | |
621 | ||
5be0389d YZ |
622 | struct ceph_aio_work { |
623 | struct work_struct work; | |
624 | struct ceph_osd_request *req; | |
625 | }; | |
626 | ||
627 | static void ceph_aio_retry_work(struct work_struct *work); | |
628 | ||
c8fe9b17 YZ |
629 | static void ceph_aio_complete(struct inode *inode, |
630 | struct ceph_aio_request *aio_req) | |
631 | { | |
632 | struct ceph_inode_info *ci = ceph_inode(inode); | |
633 | int ret; | |
634 | ||
635 | if (!atomic_dec_and_test(&aio_req->pending_reqs)) | |
636 | return; | |
637 | ||
638 | ret = aio_req->error; | |
639 | if (!ret) | |
640 | ret = aio_req->total_len; | |
641 | ||
642 | dout("ceph_aio_complete %p rc %d\n", inode, ret); | |
643 | ||
644 | if (ret >= 0 && aio_req->write) { | |
645 | int dirty; | |
646 | ||
647 | loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len; | |
648 | if (endoff > i_size_read(inode)) { | |
649 | if (ceph_inode_set_size(inode, endoff)) | |
650 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); | |
651 | } | |
652 | ||
653 | spin_lock(&ci->i_ceph_lock); | |
654 | ci->i_inline_version = CEPH_INLINE_NONE; | |
655 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, | |
656 | &aio_req->prealloc_cf); | |
657 | spin_unlock(&ci->i_ceph_lock); | |
658 | if (dirty) | |
659 | __mark_inode_dirty(inode, dirty); | |
660 | ||
661 | } | |
662 | ||
663 | ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR : | |
664 | CEPH_CAP_FILE_RD)); | |
665 | ||
666 | aio_req->iocb->ki_complete(aio_req->iocb, ret, 0); | |
667 | ||
668 | ceph_free_cap_flush(aio_req->prealloc_cf); | |
669 | kfree(aio_req); | |
670 | } | |
671 | ||
85e084fe | 672 | static void ceph_aio_complete_req(struct ceph_osd_request *req) |
c8fe9b17 YZ |
673 | { |
674 | int rc = req->r_result; | |
675 | struct inode *inode = req->r_inode; | |
676 | struct ceph_aio_request *aio_req = req->r_priv; | |
677 | struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0); | |
678 | int num_pages = calc_pages_for((u64)osd_data->alignment, | |
679 | osd_data->length); | |
680 | ||
681 | dout("ceph_aio_complete_req %p rc %d bytes %llu\n", | |
682 | inode, rc, osd_data->length); | |
683 | ||
684 | if (rc == -EOLDSNAPC) { | |
5be0389d YZ |
685 | struct ceph_aio_work *aio_work; |
686 | BUG_ON(!aio_req->write); | |
687 | ||
688 | aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS); | |
689 | if (aio_work) { | |
690 | INIT_WORK(&aio_work->work, ceph_aio_retry_work); | |
691 | aio_work->req = req; | |
692 | queue_work(ceph_inode_to_client(inode)->wb_wq, | |
693 | &aio_work->work); | |
694 | return; | |
695 | } | |
696 | rc = -ENOMEM; | |
697 | } else if (!aio_req->write) { | |
c8fe9b17 YZ |
698 | if (rc == -ENOENT) |
699 | rc = 0; | |
700 | if (rc >= 0 && osd_data->length > rc) { | |
701 | int zoff = osd_data->alignment + rc; | |
702 | int zlen = osd_data->length - rc; | |
703 | /* | |
704 | * If read is satisfied by single OSD request, | |
705 | * it can pass EOF. Otherwise read is within | |
706 | * i_size. | |
707 | */ | |
708 | if (aio_req->num_reqs == 1) { | |
709 | loff_t i_size = i_size_read(inode); | |
710 | loff_t endoff = aio_req->iocb->ki_pos + rc; | |
711 | if (endoff < i_size) | |
712 | zlen = min_t(size_t, zlen, | |
713 | i_size - endoff); | |
714 | aio_req->total_len = rc + zlen; | |
715 | } | |
716 | ||
717 | if (zlen > 0) | |
718 | ceph_zero_page_vector_range(zoff, zlen, | |
719 | osd_data->pages); | |
720 | } | |
721 | } | |
722 | ||
723 | ceph_put_page_vector(osd_data->pages, num_pages, false); | |
724 | ceph_osdc_put_request(req); | |
725 | ||
726 | if (rc < 0) | |
727 | cmpxchg(&aio_req->error, 0, rc); | |
728 | ||
729 | ceph_aio_complete(inode, aio_req); | |
730 | return; | |
731 | } | |
732 | ||
5be0389d YZ |
733 | static void ceph_aio_retry_work(struct work_struct *work) |
734 | { | |
735 | struct ceph_aio_work *aio_work = | |
736 | container_of(work, struct ceph_aio_work, work); | |
737 | struct ceph_osd_request *orig_req = aio_work->req; | |
738 | struct ceph_aio_request *aio_req = orig_req->r_priv; | |
739 | struct inode *inode = orig_req->r_inode; | |
740 | struct ceph_inode_info *ci = ceph_inode(inode); | |
741 | struct ceph_snap_context *snapc; | |
742 | struct ceph_osd_request *req; | |
743 | int ret; | |
744 | ||
745 | spin_lock(&ci->i_ceph_lock); | |
746 | if (__ceph_have_pending_cap_snap(ci)) { | |
747 | struct ceph_cap_snap *capsnap = | |
748 | list_last_entry(&ci->i_cap_snaps, | |
749 | struct ceph_cap_snap, | |
750 | ci_item); | |
751 | snapc = ceph_get_snap_context(capsnap->context); | |
752 | } else { | |
753 | BUG_ON(!ci->i_head_snapc); | |
754 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
755 | } | |
756 | spin_unlock(&ci->i_ceph_lock); | |
757 | ||
758 | req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 2, | |
759 | false, GFP_NOFS); | |
1418bf07 DC |
760 | if (!req) { |
761 | ret = -ENOMEM; | |
5be0389d YZ |
762 | req = orig_req; |
763 | goto out; | |
764 | } | |
765 | ||
766 | req->r_flags = CEPH_OSD_FLAG_ORDERSNAP | | |
767 | CEPH_OSD_FLAG_ONDISK | | |
768 | CEPH_OSD_FLAG_WRITE; | |
63244fa1 | 769 | ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc); |
d30291b9 | 770 | ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid); |
5be0389d | 771 | |
13d1ad16 ID |
772 | ret = ceph_osdc_alloc_messages(req, GFP_NOFS); |
773 | if (ret) { | |
774 | ceph_osdc_put_request(req); | |
775 | req = orig_req; | |
776 | goto out; | |
777 | } | |
5be0389d YZ |
778 | |
779 | req->r_ops[0] = orig_req->r_ops[0]; | |
780 | osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0); | |
781 | ||
bb873b53 ID |
782 | req->r_mtime = aio_req->mtime; |
783 | req->r_data_offset = req->r_ops[0].extent.offset; | |
5be0389d | 784 | |
5be0389d YZ |
785 | ceph_osdc_put_request(orig_req); |
786 | ||
787 | req->r_callback = ceph_aio_complete_req; | |
788 | req->r_inode = inode; | |
789 | req->r_priv = aio_req; | |
790 | ||
791 | ret = ceph_osdc_start_request(req->r_osdc, req, false); | |
792 | out: | |
793 | if (ret < 0) { | |
5be0389d | 794 | req->r_result = ret; |
85e084fe | 795 | ceph_aio_complete_req(req); |
5be0389d YZ |
796 | } |
797 | ||
db6aed70 | 798 | ceph_put_snap_context(snapc); |
5be0389d YZ |
799 | kfree(aio_work); |
800 | } | |
801 | ||
124e68e7 | 802 | /* |
26be8808 AE |
803 | * Write commit request unsafe callback, called to tell us when a |
804 | * request is unsafe (that is, in flight--has been handed to the | |
805 | * messenger to send to its target osd). It is called again when | |
806 | * we've received a response message indicating the request is | |
807 | * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request | |
808 | * is completed early (and unsuccessfully) due to a timeout or | |
809 | * interrupt. | |
810 | * | |
811 | * This is used if we requested both an ACK and ONDISK commit reply | |
812 | * from the OSD. | |
124e68e7 | 813 | */ |
26be8808 | 814 | static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe) |
124e68e7 SW |
815 | { |
816 | struct ceph_inode_info *ci = ceph_inode(req->r_inode); | |
817 | ||
26be8808 AE |
818 | dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid, |
819 | unsafe ? "un" : ""); | |
820 | if (unsafe) { | |
821 | ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR); | |
822 | spin_lock(&ci->i_unsafe_lock); | |
823 | list_add_tail(&req->r_unsafe_item, | |
824 | &ci->i_unsafe_writes); | |
825 | spin_unlock(&ci->i_unsafe_lock); | |
fe5da05e ID |
826 | |
827 | complete_all(&req->r_completion); | |
26be8808 AE |
828 | } else { |
829 | spin_lock(&ci->i_unsafe_lock); | |
830 | list_del_init(&req->r_unsafe_item); | |
831 | spin_unlock(&ci->i_unsafe_lock); | |
832 | ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR); | |
833 | } | |
124e68e7 SW |
834 | } |
835 | ||
e8344e66 | 836 | |
e8344e66 | 837 | static ssize_t |
c8fe9b17 YZ |
838 | ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter, |
839 | struct ceph_snap_context *snapc, | |
840 | struct ceph_cap_flush **pcf) | |
124e68e7 | 841 | { |
e8344e66 | 842 | struct file *file = iocb->ki_filp; |
496ad9aa | 843 | struct inode *inode = file_inode(file); |
124e68e7 | 844 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 | 845 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
acead002 | 846 | struct ceph_vino vino; |
124e68e7 SW |
847 | struct ceph_osd_request *req; |
848 | struct page **pages; | |
c8fe9b17 YZ |
849 | struct ceph_aio_request *aio_req = NULL; |
850 | int num_pages = 0; | |
124e68e7 | 851 | int flags; |
124e68e7 | 852 | int ret; |
8bbd4714 | 853 | struct timespec mtime = current_fs_time(inode->i_sb); |
c8fe9b17 YZ |
854 | size_t count = iov_iter_count(iter); |
855 | loff_t pos = iocb->ki_pos; | |
856 | bool write = iov_iter_rw(iter) == WRITE; | |
124e68e7 | 857 | |
c8fe9b17 | 858 | if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
124e68e7 SW |
859 | return -EROFS; |
860 | ||
c8fe9b17 YZ |
861 | dout("sync_direct_read_write (%s) on file %p %lld~%u\n", |
862 | (write ? "write" : "read"), file, pos, (unsigned)count); | |
124e68e7 | 863 | |
e8344e66 | 864 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); |
29065a51 YS |
865 | if (ret < 0) |
866 | return ret; | |
867 | ||
c8fe9b17 YZ |
868 | if (write) { |
869 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
09cbfeaf KS |
870 | pos >> PAGE_SHIFT, |
871 | (pos + count) >> PAGE_SHIFT); | |
c8fe9b17 YZ |
872 | if (ret < 0) |
873 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
29065a51 | 874 | |
c8fe9b17 YZ |
875 | flags = CEPH_OSD_FLAG_ORDERSNAP | |
876 | CEPH_OSD_FLAG_ONDISK | | |
877 | CEPH_OSD_FLAG_WRITE; | |
878 | } else { | |
879 | flags = CEPH_OSD_FLAG_READ; | |
880 | } | |
124e68e7 | 881 | |
c8fe9b17 YZ |
882 | while (iov_iter_count(iter) > 0) { |
883 | u64 size = dio_get_pagev_size(iter); | |
884 | size_t start = 0; | |
885 | ssize_t len; | |
e8344e66 | 886 | |
e8344e66 | 887 | vino = ceph_vino(inode); |
888 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
c8fe9b17 YZ |
889 | vino, pos, &size, 0, |
890 | /*include a 'startsync' command*/ | |
891 | write ? 2 : 1, | |
892 | write ? CEPH_OSD_OP_WRITE : | |
893 | CEPH_OSD_OP_READ, | |
894 | flags, snapc, | |
e8344e66 | 895 | ci->i_truncate_seq, |
896 | ci->i_truncate_size, | |
897 | false); | |
898 | if (IS_ERR(req)) { | |
899 | ret = PTR_ERR(req); | |
eab87235 | 900 | break; |
e8344e66 | 901 | } |
124e68e7 | 902 | |
c8fe9b17 YZ |
903 | len = size; |
904 | pages = dio_get_pages_alloc(iter, len, &start, &num_pages); | |
b5b98989 | 905 | if (IS_ERR(pages)) { |
64c31311 | 906 | ceph_osdc_put_request(req); |
b5b98989 | 907 | ret = PTR_ERR(pages); |
64c31311 | 908 | break; |
124e68e7 SW |
909 | } |
910 | ||
911 | /* | |
c8fe9b17 YZ |
912 | * To simplify error handling, allow AIO when IO within i_size |
913 | * or IO can be satisfied by single OSD request. | |
124e68e7 | 914 | */ |
c8fe9b17 YZ |
915 | if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) && |
916 | (len == count || pos + count <= i_size_read(inode))) { | |
917 | aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL); | |
918 | if (aio_req) { | |
919 | aio_req->iocb = iocb; | |
920 | aio_req->write = write; | |
921 | INIT_LIST_HEAD(&aio_req->osd_reqs); | |
922 | if (write) { | |
5be0389d | 923 | aio_req->mtime = mtime; |
c8fe9b17 YZ |
924 | swap(aio_req->prealloc_cf, *pcf); |
925 | } | |
926 | } | |
927 | /* ignore error */ | |
928 | } | |
929 | ||
930 | if (write) { | |
931 | /* | |
932 | * throw out any page cache pages in this range. this | |
933 | * may block. | |
934 | */ | |
935 | truncate_inode_pages_range(inode->i_mapping, pos, | |
09cbfeaf | 936 | (pos+len) | (PAGE_SIZE - 1)); |
c8fe9b17 YZ |
937 | |
938 | osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0); | |
bb873b53 | 939 | req->r_mtime = mtime; |
c8fe9b17 YZ |
940 | } |
941 | ||
c8fe9b17 YZ |
942 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, start, |
943 | false, false); | |
e8344e66 | 944 | |
c8fe9b17 YZ |
945 | if (aio_req) { |
946 | aio_req->total_len += len; | |
947 | aio_req->num_reqs++; | |
948 | atomic_inc(&aio_req->pending_reqs); | |
949 | ||
950 | req->r_callback = ceph_aio_complete_req; | |
951 | req->r_inode = inode; | |
952 | req->r_priv = aio_req; | |
953 | list_add_tail(&req->r_unsafe_item, &aio_req->osd_reqs); | |
954 | ||
955 | pos += len; | |
956 | iov_iter_advance(iter, len); | |
957 | continue; | |
958 | } | |
959 | ||
960 | ret = ceph_osdc_start_request(req->r_osdc, req, false); | |
e8344e66 | 961 | if (!ret) |
962 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
963 | ||
c8fe9b17 YZ |
964 | size = i_size_read(inode); |
965 | if (!write) { | |
966 | if (ret == -ENOENT) | |
967 | ret = 0; | |
968 | if (ret >= 0 && ret < len && pos + ret < size) { | |
969 | int zlen = min_t(size_t, len - ret, | |
970 | size - pos - ret); | |
971 | ceph_zero_page_vector_range(start + ret, zlen, | |
972 | pages); | |
973 | ret += zlen; | |
974 | } | |
975 | if (ret >= 0) | |
976 | len = ret; | |
977 | } | |
978 | ||
e8344e66 | 979 | ceph_put_page_vector(pages, num_pages, false); |
980 | ||
e8344e66 | 981 | ceph_osdc_put_request(req); |
c8fe9b17 | 982 | if (ret < 0) |
e8344e66 | 983 | break; |
64c31311 | 984 | |
c8fe9b17 YZ |
985 | pos += len; |
986 | iov_iter_advance(iter, len); | |
987 | ||
988 | if (!write && pos >= size) | |
e8344e66 | 989 | break; |
64c31311 | 990 | |
c8fe9b17 YZ |
991 | if (write && pos > size) { |
992 | if (ceph_inode_set_size(inode, pos)) | |
64c31311 AV |
993 | ceph_check_caps(ceph_inode(inode), |
994 | CHECK_CAPS_AUTHONLY, | |
995 | NULL); | |
996 | } | |
e8344e66 | 997 | } |
998 | ||
c8fe9b17 YZ |
999 | if (aio_req) { |
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 | ||
1008 | while (!list_empty(&aio_req->osd_reqs)) { | |
1009 | req = list_first_entry(&aio_req->osd_reqs, | |
1010 | struct ceph_osd_request, | |
1011 | r_unsafe_item); | |
1012 | list_del_init(&req->r_unsafe_item); | |
1013 | if (ret >= 0) | |
1014 | ret = ceph_osdc_start_request(req->r_osdc, | |
1015 | req, false); | |
1016 | if (ret < 0) { | |
1017 | req->r_result = ret; | |
85e084fe | 1018 | ceph_aio_complete_req(req); |
c8fe9b17 YZ |
1019 | } |
1020 | } | |
1021 | return -EIOCBQUEUED; | |
1022 | } | |
1023 | ||
1024 | if (ret != -EOLDSNAPC && pos > iocb->ki_pos) { | |
1025 | ret = pos - iocb->ki_pos; | |
e8344e66 | 1026 | iocb->ki_pos = pos; |
e8344e66 | 1027 | } |
1028 | return ret; | |
1029 | } | |
1030 | ||
e8344e66 | 1031 | /* |
1032 | * Synchronous write, straight from __user pointer or user pages. | |
1033 | * | |
1034 | * If write spans object boundary, just do multiple writes. (For a | |
1035 | * correct atomic write, we should e.g. take write locks on all | |
1036 | * objects, rollback on failure, etc.) | |
1037 | */ | |
06fee30f | 1038 | static ssize_t |
5dda377c YZ |
1039 | ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos, |
1040 | struct ceph_snap_context *snapc) | |
e8344e66 | 1041 | { |
1042 | struct file *file = iocb->ki_filp; | |
1043 | struct inode *inode = file_inode(file); | |
1044 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1045 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
e8344e66 | 1046 | struct ceph_vino vino; |
1047 | struct ceph_osd_request *req; | |
1048 | struct page **pages; | |
1049 | u64 len; | |
1050 | int num_pages; | |
1051 | int written = 0; | |
1052 | int flags; | |
1053 | int check_caps = 0; | |
1054 | int ret; | |
8bbd4714 | 1055 | struct timespec mtime = current_fs_time(inode->i_sb); |
4908b822 | 1056 | size_t count = iov_iter_count(from); |
e8344e66 | 1057 | |
1058 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) | |
1059 | return -EROFS; | |
1060 | ||
1061 | dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count); | |
1062 | ||
1063 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); | |
1064 | if (ret < 0) | |
1065 | return ret; | |
1066 | ||
1067 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
09cbfeaf KS |
1068 | pos >> PAGE_SHIFT, |
1069 | (pos + count) >> PAGE_SHIFT); | |
e8344e66 | 1070 | if (ret < 0) |
1071 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
1072 | ||
1073 | flags = CEPH_OSD_FLAG_ORDERSNAP | | |
1074 | CEPH_OSD_FLAG_ONDISK | | |
1075 | CEPH_OSD_FLAG_WRITE | | |
1076 | CEPH_OSD_FLAG_ACK; | |
1077 | ||
4908b822 | 1078 | while ((len = iov_iter_count(from)) > 0) { |
e8344e66 | 1079 | size_t left; |
1080 | int n; | |
1081 | ||
e8344e66 | 1082 | vino = ceph_vino(inode); |
1083 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
715e4cd4 | 1084 | vino, pos, &len, 0, 1, |
e8344e66 | 1085 | CEPH_OSD_OP_WRITE, flags, snapc, |
1086 | ci->i_truncate_seq, | |
1087 | ci->i_truncate_size, | |
1088 | false); | |
1089 | if (IS_ERR(req)) { | |
1090 | ret = PTR_ERR(req); | |
eab87235 | 1091 | break; |
e8344e66 | 1092 | } |
1093 | ||
1094 | /* | |
1095 | * write from beginning of first page, | |
1096 | * regardless of io alignment | |
1097 | */ | |
09cbfeaf | 1098 | num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
e8344e66 | 1099 | |
687265e5 | 1100 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
124e68e7 SW |
1101 | if (IS_ERR(pages)) { |
1102 | ret = PTR_ERR(pages); | |
1103 | goto out; | |
1104 | } | |
e8344e66 | 1105 | |
1106 | left = len; | |
1107 | for (n = 0; n < num_pages; n++) { | |
125d725c | 1108 | size_t plen = min_t(size_t, left, PAGE_SIZE); |
4908b822 | 1109 | ret = copy_page_from_iter(pages[n], 0, plen, from); |
e8344e66 | 1110 | if (ret != plen) { |
1111 | ret = -EFAULT; | |
1112 | break; | |
1113 | } | |
1114 | left -= ret; | |
e8344e66 | 1115 | } |
1116 | ||
124e68e7 SW |
1117 | if (ret < 0) { |
1118 | ceph_release_page_vector(pages, num_pages); | |
1119 | goto out; | |
1120 | } | |
1121 | ||
e8344e66 | 1122 | /* get a second commit callback */ |
1123 | req->r_unsafe_callback = ceph_sync_write_unsafe; | |
1124 | req->r_inode = inode; | |
124e68e7 | 1125 | |
e8344e66 | 1126 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, |
1127 | false, true); | |
02ee07d3 | 1128 | |
bb873b53 | 1129 | req->r_mtime = mtime; |
e8344e66 | 1130 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
1131 | if (!ret) | |
1132 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
124e68e7 SW |
1133 | |
1134 | out: | |
e8344e66 | 1135 | ceph_osdc_put_request(req); |
1136 | if (ret == 0) { | |
1137 | pos += len; | |
1138 | written += len; | |
1139 | ||
1140 | if (pos > i_size_read(inode)) { | |
1141 | check_caps = ceph_inode_set_size(inode, pos); | |
1142 | if (check_caps) | |
1143 | ceph_check_caps(ceph_inode(inode), | |
1144 | CHECK_CAPS_AUTHONLY, | |
1145 | NULL); | |
1146 | } | |
1147 | } else | |
1148 | break; | |
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 { |
8eb4efb0 | 1210 | dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n", |
3644424d | 1211 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, |
8eb4efb0 | 1212 | ceph_cap_string(got)); |
124e68e7 | 1213 | |
3644424d | 1214 | ret = generic_file_read_iter(iocb, to); |
8eb4efb0 | 1215 | } |
124e68e7 SW |
1216 | dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", |
1217 | inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); | |
3738daa6 | 1218 | if (pinned_page) { |
09cbfeaf | 1219 | put_page(pinned_page); |
3738daa6 YZ |
1220 | pinned_page = NULL; |
1221 | } | |
124e68e7 | 1222 | ceph_put_cap_refs(ci, got); |
c8fe9b17 | 1223 | if (retry_op > HAVE_RETRIED && ret >= 0) { |
83701246 YZ |
1224 | int statret; |
1225 | struct page *page = NULL; | |
1226 | loff_t i_size; | |
1227 | if (retry_op == READ_INLINE) { | |
687265e5 | 1228 | page = __page_cache_alloc(GFP_KERNEL); |
83701246 YZ |
1229 | if (!page) |
1230 | return -ENOMEM; | |
1231 | } | |
6a026589 | 1232 | |
83701246 YZ |
1233 | statret = __ceph_do_getattr(inode, page, |
1234 | CEPH_STAT_CAP_INLINE_DATA, !!page); | |
1235 | if (statret < 0) { | |
1236 | __free_page(page); | |
1237 | if (statret == -ENODATA) { | |
1238 | BUG_ON(retry_op != READ_INLINE); | |
1239 | goto again; | |
1240 | } | |
1241 | return statret; | |
1242 | } | |
6a026589 | 1243 | |
83701246 YZ |
1244 | i_size = i_size_read(inode); |
1245 | if (retry_op == READ_INLINE) { | |
fcc02d2a YZ |
1246 | BUG_ON(ret > 0 || read > 0); |
1247 | if (iocb->ki_pos < i_size && | |
09cbfeaf | 1248 | iocb->ki_pos < PAGE_SIZE) { |
83701246 YZ |
1249 | loff_t end = min_t(loff_t, i_size, |
1250 | iocb->ki_pos + len); | |
09cbfeaf | 1251 | end = min_t(loff_t, end, PAGE_SIZE); |
83701246 YZ |
1252 | if (statret < end) |
1253 | zero_user_segment(page, statret, end); | |
1254 | ret = copy_page_to_iter(page, | |
1255 | iocb->ki_pos & ~PAGE_MASK, | |
1256 | end - iocb->ki_pos, to); | |
1257 | iocb->ki_pos += ret; | |
fcc02d2a YZ |
1258 | read += ret; |
1259 | } | |
1260 | if (iocb->ki_pos < i_size && read < len) { | |
1261 | size_t zlen = min_t(size_t, len - read, | |
1262 | i_size - iocb->ki_pos); | |
1263 | ret = iov_iter_zero(zlen, to); | |
1264 | iocb->ki_pos += ret; | |
1265 | read += ret; | |
83701246 YZ |
1266 | } |
1267 | __free_pages(page, 0); | |
fcc02d2a | 1268 | return read; |
83701246 | 1269 | } |
6a026589 SW |
1270 | |
1271 | /* hit EOF or hole? */ | |
83701246 | 1272 | if (retry_op == CHECK_EOF && iocb->ki_pos < i_size && |
fcc02d2a | 1273 | ret < len) { |
8eb4efb0 | 1274 | dout("sync_read hit hole, ppos %lld < size %lld" |
99c88e69 | 1275 | ", reading more\n", iocb->ki_pos, i_size); |
8eb4efb0 | 1276 | |
6a026589 | 1277 | read += ret; |
6a026589 | 1278 | len -= ret; |
c8fe9b17 | 1279 | retry_op = HAVE_RETRIED; |
6a026589 SW |
1280 | goto again; |
1281 | } | |
1282 | } | |
8eb4efb0 | 1283 | |
6a026589 SW |
1284 | if (ret >= 0) |
1285 | ret += read; | |
1286 | ||
124e68e7 SW |
1287 | return ret; |
1288 | } | |
1289 | ||
1290 | /* | |
1291 | * Take cap references to avoid releasing caps to MDS mid-write. | |
1292 | * | |
1293 | * If we are synchronous, and write with an old snap context, the OSD | |
1294 | * may return EOLDSNAPC. In that case, retry the write.. _after_ | |
1295 | * dropping our cap refs and allowing the pending snap to logically | |
1296 | * complete _before_ this write occurs. | |
1297 | * | |
1298 | * If we are near ENOSPC, write synchronously. | |
1299 | */ | |
4908b822 | 1300 | static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from) |
124e68e7 SW |
1301 | { |
1302 | struct file *file = iocb->ki_filp; | |
33caad32 | 1303 | struct ceph_file_info *fi = file->private_data; |
496ad9aa | 1304 | struct inode *inode = file_inode(file); |
124e68e7 | 1305 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 YS |
1306 | struct ceph_osd_client *osdc = |
1307 | &ceph_sb_to_client(inode->i_sb)->client->osdc; | |
f66fd9f0 | 1308 | struct ceph_cap_flush *prealloc_cf; |
3309dd04 | 1309 | ssize_t count, written = 0; |
03d254ed | 1310 | int err, want, got; |
3309dd04 | 1311 | loff_t pos; |
124e68e7 SW |
1312 | |
1313 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
1314 | return -EROFS; | |
1315 | ||
f66fd9f0 YZ |
1316 | prealloc_cf = ceph_alloc_cap_flush(); |
1317 | if (!prealloc_cf) | |
1318 | return -ENOMEM; | |
1319 | ||
5955102c | 1320 | inode_lock(inode); |
03d254ed | 1321 | |
03d254ed | 1322 | /* We can write back this queue in page reclaim */ |
de1414a6 | 1323 | current->backing_dev_info = inode_to_bdi(inode); |
03d254ed | 1324 | |
55b0b31c YZ |
1325 | if (iocb->ki_flags & IOCB_APPEND) { |
1326 | err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); | |
1327 | if (err < 0) | |
1328 | goto out; | |
1329 | } | |
1330 | ||
3309dd04 AV |
1331 | err = generic_write_checks(iocb, from); |
1332 | if (err <= 0) | |
03d254ed YZ |
1333 | goto out; |
1334 | ||
3309dd04 AV |
1335 | pos = iocb->ki_pos; |
1336 | count = iov_iter_count(from); | |
5fa8e0a1 | 1337 | err = file_remove_privs(file); |
03d254ed YZ |
1338 | if (err) |
1339 | goto out; | |
1340 | ||
1341 | err = file_update_time(file); | |
1342 | if (err) | |
1343 | goto out; | |
1344 | ||
28127bdd YZ |
1345 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
1346 | err = ceph_uninline_data(file, NULL); | |
1347 | if (err < 0) | |
1348 | goto out; | |
1349 | } | |
1350 | ||
124e68e7 | 1351 | retry_snap: |
6070e0c1 | 1352 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) { |
03d254ed | 1353 | err = -ENOSPC; |
6070e0c1 YZ |
1354 | goto out; |
1355 | } | |
03d254ed | 1356 | |
ac7f29bf | 1357 | dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n", |
99c88e69 | 1358 | inode, ceph_vinop(inode), pos, count, i_size_read(inode)); |
7971bd92 SW |
1359 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
1360 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1361 | else | |
1362 | want = CEPH_CAP_FILE_BUFFER; | |
03d254ed | 1363 | got = 0; |
3738daa6 YZ |
1364 | err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count, |
1365 | &got, NULL); | |
03d254ed | 1366 | if (err < 0) |
37505d57 | 1367 | goto out; |
124e68e7 | 1368 | |
ac7f29bf | 1369 | dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n", |
03d254ed | 1370 | inode, ceph_vinop(inode), pos, count, ceph_cap_string(got)); |
7971bd92 SW |
1371 | |
1372 | if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || | |
2ba48ce5 | 1373 | (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC)) { |
5dda377c | 1374 | struct ceph_snap_context *snapc; |
4908b822 | 1375 | struct iov_iter data; |
5955102c | 1376 | inode_unlock(inode); |
5dda377c YZ |
1377 | |
1378 | spin_lock(&ci->i_ceph_lock); | |
1379 | if (__ceph_have_pending_cap_snap(ci)) { | |
1380 | struct ceph_cap_snap *capsnap = | |
1381 | list_last_entry(&ci->i_cap_snaps, | |
1382 | struct ceph_cap_snap, | |
1383 | ci_item); | |
1384 | snapc = ceph_get_snap_context(capsnap->context); | |
1385 | } else { | |
1386 | BUG_ON(!ci->i_head_snapc); | |
1387 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
1388 | } | |
1389 | spin_unlock(&ci->i_ceph_lock); | |
1390 | ||
4908b822 AV |
1391 | /* we might need to revert back to that point */ |
1392 | data = *from; | |
2ba48ce5 | 1393 | if (iocb->ki_flags & IOCB_DIRECT) |
c8fe9b17 YZ |
1394 | written = ceph_direct_read_write(iocb, &data, snapc, |
1395 | &prealloc_cf); | |
e8344e66 | 1396 | else |
5dda377c | 1397 | written = ceph_sync_write(iocb, &data, pos, snapc); |
0e5dd45c | 1398 | if (written == -EOLDSNAPC) { |
1399 | dout("aio_write %p %llx.%llx %llu~%u" | |
1400 | "got EOLDSNAPC, retrying\n", | |
1401 | inode, ceph_vinop(inode), | |
4908b822 | 1402 | pos, (unsigned)count); |
5955102c | 1403 | inode_lock(inode); |
0e5dd45c | 1404 | goto retry_snap; |
1405 | } | |
4908b822 AV |
1406 | if (written > 0) |
1407 | iov_iter_advance(from, written); | |
5dda377c | 1408 | ceph_put_snap_context(snapc); |
7971bd92 | 1409 | } else { |
99c88e69 | 1410 | loff_t old_size = i_size_read(inode); |
b0d7c223 YZ |
1411 | /* |
1412 | * No need to acquire the i_truncate_mutex. Because | |
1413 | * the MDS revokes Fwb caps before sending truncate | |
1414 | * message to us. We can't get Fwb cap while there | |
1415 | * are pending vmtruncate. So write and vmtruncate | |
1416 | * can not run at the same time | |
1417 | */ | |
4908b822 | 1418 | written = generic_perform_write(file, from, pos); |
aec605f4 AV |
1419 | if (likely(written >= 0)) |
1420 | iocb->ki_pos = pos + written; | |
99c88e69 | 1421 | if (i_size_read(inode) > old_size) |
32d3e148 | 1422 | ceph_fscache_update_objectsize(inode); |
5955102c | 1423 | inode_unlock(inode); |
7971bd92 | 1424 | } |
d8de9ab6 | 1425 | |
03d254ed | 1426 | if (written >= 0) { |
fca65b4a | 1427 | int dirty; |
be655596 | 1428 | spin_lock(&ci->i_ceph_lock); |
28127bdd | 1429 | ci->i_inline_version = CEPH_INLINE_NONE; |
f66fd9f0 YZ |
1430 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
1431 | &prealloc_cf); | |
be655596 | 1432 | spin_unlock(&ci->i_ceph_lock); |
fca65b4a SW |
1433 | if (dirty) |
1434 | __mark_inode_dirty(inode, dirty); | |
124e68e7 | 1435 | } |
7971bd92 | 1436 | |
124e68e7 | 1437 | dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n", |
4908b822 | 1438 | inode, ceph_vinop(inode), pos, (unsigned)count, |
7971bd92 | 1439 | ceph_cap_string(got)); |
124e68e7 | 1440 | ceph_put_cap_refs(ci, got); |
7971bd92 | 1441 | |
6aa657c8 CH |
1442 | if (written >= 0) { |
1443 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL)) | |
1444 | iocb->ki_flags |= IOCB_DSYNC; | |
1445 | ||
1446 | written = generic_write_sync(iocb, written); | |
6070e0c1 | 1447 | } |
03d254ed | 1448 | |
2f75e9e1 SW |
1449 | goto out_unlocked; |
1450 | ||
03d254ed | 1451 | out: |
5955102c | 1452 | inode_unlock(inode); |
2f75e9e1 | 1453 | out_unlocked: |
f66fd9f0 | 1454 | ceph_free_cap_flush(prealloc_cf); |
03d254ed | 1455 | current->backing_dev_info = NULL; |
03d254ed | 1456 | return written ? written : err; |
124e68e7 SW |
1457 | } |
1458 | ||
1459 | /* | |
1460 | * llseek. be sure to verify file size on SEEK_END. | |
1461 | */ | |
965c8e59 | 1462 | static loff_t ceph_llseek(struct file *file, loff_t offset, int whence) |
124e68e7 SW |
1463 | { |
1464 | struct inode *inode = file->f_mapping->host; | |
99c88e69 | 1465 | loff_t i_size; |
124e68e7 SW |
1466 | int ret; |
1467 | ||
5955102c | 1468 | inode_lock(inode); |
6a82c47a | 1469 | |
965c8e59 | 1470 | if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { |
508b32d8 | 1471 | ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); |
124e68e7 SW |
1472 | if (ret < 0) { |
1473 | offset = ret; | |
1474 | goto out; | |
1475 | } | |
06222e49 JB |
1476 | } |
1477 | ||
99c88e69 | 1478 | i_size = i_size_read(inode); |
965c8e59 | 1479 | switch (whence) { |
06222e49 | 1480 | case SEEK_END: |
99c88e69 | 1481 | offset += i_size; |
124e68e7 SW |
1482 | break; |
1483 | case SEEK_CUR: | |
1484 | /* | |
1485 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
1486 | * position-querying operation. Avoid rewriting the "same" | |
1487 | * f_pos value back to the file because a concurrent read(), | |
1488 | * write() or lseek() might have altered it | |
1489 | */ | |
1490 | if (offset == 0) { | |
1491 | offset = file->f_pos; | |
1492 | goto out; | |
1493 | } | |
1494 | offset += file->f_pos; | |
1495 | break; | |
06222e49 | 1496 | case SEEK_DATA: |
99c88e69 | 1497 | if (offset >= i_size) { |
06222e49 JB |
1498 | ret = -ENXIO; |
1499 | goto out; | |
1500 | } | |
1501 | break; | |
1502 | case SEEK_HOLE: | |
99c88e69 | 1503 | if (offset >= i_size) { |
06222e49 JB |
1504 | ret = -ENXIO; |
1505 | goto out; | |
1506 | } | |
99c88e69 | 1507 | offset = i_size; |
06222e49 | 1508 | break; |
124e68e7 SW |
1509 | } |
1510 | ||
46a1c2c7 | 1511 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
124e68e7 SW |
1512 | |
1513 | out: | |
5955102c | 1514 | inode_unlock(inode); |
124e68e7 SW |
1515 | return offset; |
1516 | } | |
1517 | ||
ad7a60de LW |
1518 | static inline void ceph_zero_partial_page( |
1519 | struct inode *inode, loff_t offset, unsigned size) | |
1520 | { | |
1521 | struct page *page; | |
09cbfeaf | 1522 | pgoff_t index = offset >> PAGE_SHIFT; |
ad7a60de LW |
1523 | |
1524 | page = find_lock_page(inode->i_mapping, index); | |
1525 | if (page) { | |
1526 | wait_on_page_writeback(page); | |
09cbfeaf | 1527 | zero_user(page, offset & (PAGE_SIZE - 1), size); |
ad7a60de | 1528 | unlock_page(page); |
09cbfeaf | 1529 | put_page(page); |
ad7a60de LW |
1530 | } |
1531 | } | |
1532 | ||
1533 | static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset, | |
1534 | loff_t length) | |
1535 | { | |
09cbfeaf | 1536 | loff_t nearly = round_up(offset, PAGE_SIZE); |
ad7a60de LW |
1537 | if (offset < nearly) { |
1538 | loff_t size = nearly - offset; | |
1539 | if (length < size) | |
1540 | size = length; | |
1541 | ceph_zero_partial_page(inode, offset, size); | |
1542 | offset += size; | |
1543 | length -= size; | |
1544 | } | |
09cbfeaf KS |
1545 | if (length >= PAGE_SIZE) { |
1546 | loff_t size = round_down(length, PAGE_SIZE); | |
ad7a60de LW |
1547 | truncate_pagecache_range(inode, offset, offset + size - 1); |
1548 | offset += size; | |
1549 | length -= size; | |
1550 | } | |
1551 | if (length) | |
1552 | ceph_zero_partial_page(inode, offset, length); | |
1553 | } | |
1554 | ||
1555 | static int ceph_zero_partial_object(struct inode *inode, | |
1556 | loff_t offset, loff_t *length) | |
1557 | { | |
1558 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1559 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
1560 | struct ceph_osd_request *req; | |
1561 | int ret = 0; | |
1562 | loff_t zero = 0; | |
1563 | int op; | |
1564 | ||
1565 | if (!length) { | |
1566 | op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE; | |
1567 | length = &zero; | |
1568 | } else { | |
1569 | op = CEPH_OSD_OP_ZERO; | |
1570 | } | |
1571 | ||
1572 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1573 | ceph_vino(inode), | |
1574 | offset, length, | |
715e4cd4 | 1575 | 0, 1, op, |
ad7a60de LW |
1576 | CEPH_OSD_FLAG_WRITE | |
1577 | CEPH_OSD_FLAG_ONDISK, | |
1578 | NULL, 0, 0, false); | |
1579 | if (IS_ERR(req)) { | |
1580 | ret = PTR_ERR(req); | |
1581 | goto out; | |
1582 | } | |
1583 | ||
bb873b53 | 1584 | req->r_mtime = inode->i_mtime; |
ad7a60de LW |
1585 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
1586 | if (!ret) { | |
1587 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1588 | if (ret == -ENOENT) | |
1589 | ret = 0; | |
1590 | } | |
1591 | ceph_osdc_put_request(req); | |
1592 | ||
1593 | out: | |
1594 | return ret; | |
1595 | } | |
1596 | ||
1597 | static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length) | |
1598 | { | |
1599 | int ret = 0; | |
1600 | struct ceph_inode_info *ci = ceph_inode(inode); | |
b314a90d SW |
1601 | s32 stripe_unit = ceph_file_layout_su(ci->i_layout); |
1602 | s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout); | |
1603 | s32 object_size = ceph_file_layout_object_size(ci->i_layout); | |
1604 | u64 object_set_size = object_size * stripe_count; | |
1605 | u64 nearly, t; | |
1606 | ||
1607 | /* round offset up to next period boundary */ | |
1608 | nearly = offset + object_set_size - 1; | |
1609 | t = nearly; | |
1610 | nearly -= do_div(t, object_set_size); | |
ad7a60de | 1611 | |
ad7a60de LW |
1612 | while (length && offset < nearly) { |
1613 | loff_t size = length; | |
1614 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1615 | if (ret < 0) | |
1616 | return ret; | |
1617 | offset += size; | |
1618 | length -= size; | |
1619 | } | |
1620 | while (length >= object_set_size) { | |
1621 | int i; | |
1622 | loff_t pos = offset; | |
1623 | for (i = 0; i < stripe_count; ++i) { | |
1624 | ret = ceph_zero_partial_object(inode, pos, NULL); | |
1625 | if (ret < 0) | |
1626 | return ret; | |
1627 | pos += stripe_unit; | |
1628 | } | |
1629 | offset += object_set_size; | |
1630 | length -= object_set_size; | |
1631 | } | |
1632 | while (length) { | |
1633 | loff_t size = length; | |
1634 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1635 | if (ret < 0) | |
1636 | return ret; | |
1637 | offset += size; | |
1638 | length -= size; | |
1639 | } | |
1640 | return ret; | |
1641 | } | |
1642 | ||
1643 | static long ceph_fallocate(struct file *file, int mode, | |
1644 | loff_t offset, loff_t length) | |
1645 | { | |
1646 | struct ceph_file_info *fi = file->private_data; | |
aa8b60e0 | 1647 | struct inode *inode = file_inode(file); |
ad7a60de LW |
1648 | struct ceph_inode_info *ci = ceph_inode(inode); |
1649 | struct ceph_osd_client *osdc = | |
1650 | &ceph_inode_to_client(inode)->client->osdc; | |
f66fd9f0 | 1651 | struct ceph_cap_flush *prealloc_cf; |
ad7a60de LW |
1652 | int want, got = 0; |
1653 | int dirty; | |
1654 | int ret = 0; | |
1655 | loff_t endoff = 0; | |
1656 | loff_t size; | |
1657 | ||
494d77bf YZ |
1658 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
1659 | return -EOPNOTSUPP; | |
1660 | ||
ad7a60de LW |
1661 | if (!S_ISREG(inode->i_mode)) |
1662 | return -EOPNOTSUPP; | |
1663 | ||
f66fd9f0 YZ |
1664 | prealloc_cf = ceph_alloc_cap_flush(); |
1665 | if (!prealloc_cf) | |
1666 | return -ENOMEM; | |
1667 | ||
5955102c | 1668 | inode_lock(inode); |
ad7a60de LW |
1669 | |
1670 | if (ceph_snap(inode) != CEPH_NOSNAP) { | |
1671 | ret = -EROFS; | |
1672 | goto unlock; | |
1673 | } | |
1674 | ||
1675 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) && | |
1676 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
1677 | ret = -ENOSPC; | |
1678 | goto unlock; | |
1679 | } | |
1680 | ||
28127bdd YZ |
1681 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
1682 | ret = ceph_uninline_data(file, NULL); | |
1683 | if (ret < 0) | |
1684 | goto unlock; | |
1685 | } | |
1686 | ||
ad7a60de LW |
1687 | size = i_size_read(inode); |
1688 | if (!(mode & FALLOC_FL_KEEP_SIZE)) | |
1689 | endoff = offset + length; | |
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, | |
124e68e7 | 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 |