Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
3d14c5d2 | 2 | #include <linux/ceph/ceph_debug.h> |
124e68e7 | 3 | |
3d14c5d2 | 4 | #include <linux/module.h> |
124e68e7 | 5 | #include <linux/sched.h> |
5a0e3ad6 | 6 | #include <linux/slab.h> |
124e68e7 | 7 | #include <linux/file.h> |
5ef50c3b | 8 | #include <linux/mount.h> |
124e68e7 SW |
9 | #include <linux/namei.h> |
10 | #include <linux/writeback.h> | |
ad7a60de | 11 | #include <linux/falloc.h> |
124e68e7 SW |
12 | |
13 | #include "super.h" | |
14 | #include "mds_client.h" | |
99ccbd22 | 15 | #include "cache.h" |
124e68e7 | 16 | |
f775ff7d AG |
17 | static __le32 ceph_flags_sys2wire(u32 flags) |
18 | { | |
19 | u32 wire_flags = 0; | |
20 | ||
21 | switch (flags & O_ACCMODE) { | |
22 | case O_RDONLY: | |
23 | wire_flags |= CEPH_O_RDONLY; | |
24 | break; | |
25 | case O_WRONLY: | |
26 | wire_flags |= CEPH_O_WRONLY; | |
27 | break; | |
28 | case O_RDWR: | |
29 | wire_flags |= CEPH_O_RDWR; | |
30 | break; | |
31 | } | |
32 | ||
51b10f3f CX |
33 | flags &= ~O_ACCMODE; |
34 | ||
f775ff7d AG |
35 | #define ceph_sys2wire(a) if (flags & a) { wire_flags |= CEPH_##a; flags &= ~a; } |
36 | ||
37 | ceph_sys2wire(O_CREAT); | |
38 | ceph_sys2wire(O_EXCL); | |
39 | ceph_sys2wire(O_TRUNC); | |
40 | ceph_sys2wire(O_DIRECTORY); | |
41 | ceph_sys2wire(O_NOFOLLOW); | |
42 | ||
43 | #undef ceph_sys2wire | |
44 | ||
45 | if (flags) | |
4c069a58 | 46 | dout("unused open flags: %x\n", flags); |
f775ff7d AG |
47 | |
48 | return cpu_to_le32(wire_flags); | |
49 | } | |
50 | ||
124e68e7 SW |
51 | /* |
52 | * Ceph file operations | |
53 | * | |
54 | * Implement basic open/close functionality, and implement | |
55 | * read/write. | |
56 | * | |
57 | * We implement three modes of file I/O: | |
58 | * - buffered uses the generic_file_aio_{read,write} helpers | |
59 | * | |
60 | * - synchronous is used when there is multi-client read/write | |
61 | * sharing, avoids the page cache, and synchronously waits for an | |
62 | * ack from the OSD. | |
63 | * | |
64 | * - direct io takes the variant of the sync path that references | |
65 | * user pages directly. | |
66 | * | |
67 | * fsync() flushes and waits on dirty pages, but just queues metadata | |
68 | * for writeback: since the MDS can recover size and mtime there is no | |
69 | * need to wait for MDS acknowledgement. | |
70 | */ | |
71 | ||
b5b98989 | 72 | /* |
fc218544 ID |
73 | * How many pages to get in one call to iov_iter_get_pages(). This |
74 | * determines the size of the on-stack array used as a buffer. | |
b5b98989 | 75 | */ |
fc218544 ID |
76 | #define ITER_GET_BVECS_PAGES 64 |
77 | ||
78 | static ssize_t __iter_get_bvecs(struct iov_iter *iter, size_t maxsize, | |
79 | struct bio_vec *bvecs) | |
b5b98989 | 80 | { |
fc218544 ID |
81 | size_t size = 0; |
82 | int bvec_idx = 0; | |
83 | ||
84 | if (maxsize > iov_iter_count(iter)) | |
85 | maxsize = iov_iter_count(iter); | |
86 | ||
87 | while (size < maxsize) { | |
88 | struct page *pages[ITER_GET_BVECS_PAGES]; | |
89 | ssize_t bytes; | |
90 | size_t start; | |
91 | int idx = 0; | |
92 | ||
93 | bytes = iov_iter_get_pages(iter, pages, maxsize - size, | |
94 | ITER_GET_BVECS_PAGES, &start); | |
95 | if (bytes < 0) | |
96 | return size ?: bytes; | |
97 | ||
98 | iov_iter_advance(iter, bytes); | |
99 | size += bytes; | |
100 | ||
101 | for ( ; bytes; idx++, bvec_idx++) { | |
102 | struct bio_vec bv = { | |
103 | .bv_page = pages[idx], | |
104 | .bv_len = min_t(int, bytes, PAGE_SIZE - start), | |
105 | .bv_offset = start, | |
106 | }; | |
107 | ||
108 | bvecs[bvec_idx] = bv; | |
109 | bytes -= bv.bv_len; | |
110 | start = 0; | |
111 | } | |
112 | } | |
113 | ||
114 | return size; | |
b5b98989 ZC |
115 | } |
116 | ||
117 | /* | |
fc218544 ID |
118 | * iov_iter_get_pages() only considers one iov_iter segment, no matter |
119 | * what maxsize or maxpages are given. For ITER_BVEC that is a single | |
120 | * page. | |
121 | * | |
122 | * Attempt to get up to @maxsize bytes worth of pages from @iter. | |
123 | * Return the number of bytes in the created bio_vec array, or an error. | |
b5b98989 | 124 | */ |
fc218544 ID |
125 | static ssize_t iter_get_bvecs_alloc(struct iov_iter *iter, size_t maxsize, |
126 | struct bio_vec **bvecs, int *num_bvecs) | |
b5b98989 | 127 | { |
fc218544 ID |
128 | struct bio_vec *bv; |
129 | size_t orig_count = iov_iter_count(iter); | |
130 | ssize_t bytes; | |
131 | int npages; | |
b5b98989 | 132 | |
fc218544 ID |
133 | iov_iter_truncate(iter, maxsize); |
134 | npages = iov_iter_npages(iter, INT_MAX); | |
135 | iov_iter_reexpand(iter, orig_count); | |
b5b98989 | 136 | |
fc218544 ID |
137 | /* |
138 | * __iter_get_bvecs() may populate only part of the array -- zero it | |
139 | * out. | |
140 | */ | |
141 | bv = kvmalloc_array(npages, sizeof(*bv), GFP_KERNEL | __GFP_ZERO); | |
142 | if (!bv) | |
143 | return -ENOMEM; | |
b5b98989 | 144 | |
fc218544 ID |
145 | bytes = __iter_get_bvecs(iter, maxsize, bv); |
146 | if (bytes < 0) { | |
147 | /* | |
148 | * No pages were pinned -- just free the array. | |
149 | */ | |
150 | kvfree(bv); | |
151 | return bytes; | |
b5b98989 ZC |
152 | } |
153 | ||
fc218544 ID |
154 | *bvecs = bv; |
155 | *num_bvecs = npages; | |
156 | return bytes; | |
157 | } | |
158 | ||
159 | static void put_bvecs(struct bio_vec *bvecs, int num_bvecs, bool should_dirty) | |
160 | { | |
161 | int i; | |
162 | ||
163 | for (i = 0; i < num_bvecs; i++) { | |
164 | if (bvecs[i].bv_page) { | |
165 | if (should_dirty) | |
166 | set_page_dirty_lock(bvecs[i].bv_page); | |
167 | put_page(bvecs[i].bv_page); | |
168 | } | |
169 | } | |
170 | kvfree(bvecs); | |
b5b98989 | 171 | } |
124e68e7 SW |
172 | |
173 | /* | |
174 | * Prepare an open request. Preallocate ceph_cap to avoid an | |
175 | * inopportune ENOMEM later. | |
176 | */ | |
177 | static struct ceph_mds_request * | |
178 | prepare_open_request(struct super_block *sb, int flags, int create_mode) | |
179 | { | |
3d14c5d2 YS |
180 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
181 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
182 | struct ceph_mds_request *req; |
183 | int want_auth = USE_ANY_MDS; | |
184 | int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN; | |
185 | ||
186 | if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC)) | |
187 | want_auth = USE_AUTH_MDS; | |
188 | ||
189 | req = ceph_mdsc_create_request(mdsc, op, want_auth); | |
190 | if (IS_ERR(req)) | |
191 | goto out; | |
192 | req->r_fmode = ceph_flags_to_mode(flags); | |
f775ff7d | 193 | req->r_args.open.flags = ceph_flags_sys2wire(flags); |
124e68e7 | 194 | req->r_args.open.mode = cpu_to_le32(create_mode); |
124e68e7 SW |
195 | out: |
196 | return req; | |
197 | } | |
198 | ||
bb48bd4d CX |
199 | static int ceph_init_file_info(struct inode *inode, struct file *file, |
200 | int fmode, bool isdir) | |
201 | { | |
202 | struct ceph_file_info *fi; | |
203 | ||
204 | dout("%s %p %p 0%o (%s)\n", __func__, inode, file, | |
205 | inode->i_mode, isdir ? "dir" : "regular"); | |
206 | BUG_ON(inode->i_fop->release != ceph_release); | |
207 | ||
208 | if (isdir) { | |
209 | struct ceph_dir_file_info *dfi = | |
210 | kmem_cache_zalloc(ceph_dir_file_cachep, GFP_KERNEL); | |
211 | if (!dfi) { | |
212 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
213 | return -ENOMEM; | |
214 | } | |
215 | ||
216 | file->private_data = dfi; | |
217 | fi = &dfi->file_info; | |
218 | dfi->next_offset = 2; | |
219 | dfi->readdir_cache_idx = -1; | |
220 | } else { | |
221 | fi = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL); | |
222 | if (!fi) { | |
223 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
224 | return -ENOMEM; | |
225 | } | |
226 | ||
227 | file->private_data = fi; | |
228 | } | |
229 | ||
230 | fi->fmode = fmode; | |
231 | spin_lock_init(&fi->rw_contexts_lock); | |
232 | INIT_LIST_HEAD(&fi->rw_contexts); | |
233 | ||
234 | return 0; | |
235 | } | |
236 | ||
124e68e7 SW |
237 | /* |
238 | * initialize private struct file data. | |
239 | * if we fail, clean up by dropping fmode reference on the ceph_inode | |
240 | */ | |
241 | static int ceph_init_file(struct inode *inode, struct file *file, int fmode) | |
242 | { | |
124e68e7 SW |
243 | int ret = 0; |
244 | ||
245 | switch (inode->i_mode & S_IFMT) { | |
246 | case S_IFREG: | |
46b59b2b YZ |
247 | ceph_fscache_register_inode_cookie(inode); |
248 | ceph_fscache_file_set_cookie(inode, file); | |
124e68e7 | 249 | case S_IFDIR: |
bb48bd4d CX |
250 | ret = ceph_init_file_info(inode, file, fmode, |
251 | S_ISDIR(inode->i_mode)); | |
252 | if (ret) | |
253 | return ret; | |
124e68e7 SW |
254 | break; |
255 | ||
256 | case S_IFLNK: | |
257 | dout("init_file %p %p 0%o (symlink)\n", inode, file, | |
258 | inode->i_mode); | |
259 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
260 | break; | |
261 | ||
262 | default: | |
263 | dout("init_file %p %p 0%o (special)\n", inode, file, | |
264 | inode->i_mode); | |
265 | /* | |
266 | * we need to drop the open ref now, since we don't | |
267 | * have .release set to ceph_release. | |
268 | */ | |
269 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
270 | BUG_ON(inode->i_fop->release == ceph_release); | |
271 | ||
272 | /* call the proper open fop */ | |
273 | ret = inode->i_fop->open(inode, file); | |
274 | } | |
275 | return ret; | |
276 | } | |
277 | ||
77310320 YZ |
278 | /* |
279 | * try renew caps after session gets killed. | |
280 | */ | |
281 | int ceph_renew_caps(struct inode *inode) | |
282 | { | |
283 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; | |
284 | struct ceph_inode_info *ci = ceph_inode(inode); | |
285 | struct ceph_mds_request *req; | |
286 | int err, flags, wanted; | |
287 | ||
288 | spin_lock(&ci->i_ceph_lock); | |
289 | wanted = __ceph_caps_file_wanted(ci); | |
290 | if (__ceph_is_any_real_caps(ci) && | |
8242c9f3 | 291 | (!(wanted & CEPH_CAP_ANY_WR) || ci->i_auth_cap)) { |
77310320 YZ |
292 | int issued = __ceph_caps_issued(ci, NULL); |
293 | spin_unlock(&ci->i_ceph_lock); | |
294 | dout("renew caps %p want %s issued %s updating mds_wanted\n", | |
295 | inode, ceph_cap_string(wanted), ceph_cap_string(issued)); | |
296 | ceph_check_caps(ci, 0, NULL); | |
297 | return 0; | |
298 | } | |
299 | spin_unlock(&ci->i_ceph_lock); | |
300 | ||
301 | flags = 0; | |
302 | if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR)) | |
303 | flags = O_RDWR; | |
304 | else if (wanted & CEPH_CAP_FILE_RD) | |
305 | flags = O_RDONLY; | |
306 | else if (wanted & CEPH_CAP_FILE_WR) | |
307 | flags = O_WRONLY; | |
308 | #ifdef O_LAZY | |
309 | if (wanted & CEPH_CAP_FILE_LAZYIO) | |
310 | flags |= O_LAZY; | |
311 | #endif | |
312 | ||
313 | req = prepare_open_request(inode->i_sb, flags, 0); | |
314 | if (IS_ERR(req)) { | |
315 | err = PTR_ERR(req); | |
316 | goto out; | |
317 | } | |
318 | ||
319 | req->r_inode = inode; | |
320 | ihold(inode); | |
321 | req->r_num_caps = 1; | |
322 | req->r_fmode = -1; | |
323 | ||
324 | err = ceph_mdsc_do_request(mdsc, NULL, req); | |
325 | ceph_mdsc_put_request(req); | |
326 | out: | |
327 | dout("renew caps %p open result=%d\n", inode, err); | |
328 | return err < 0 ? err : 0; | |
329 | } | |
330 | ||
124e68e7 | 331 | /* |
124e68e7 SW |
332 | * If we already have the requisite capabilities, we can satisfy |
333 | * the open request locally (no need to request new caps from the | |
334 | * MDS). We do, however, need to inform the MDS (asynchronously) | |
335 | * if our wanted caps set expands. | |
336 | */ | |
337 | int ceph_open(struct inode *inode, struct file *file) | |
338 | { | |
339 | struct ceph_inode_info *ci = ceph_inode(inode); | |
3d14c5d2 YS |
340 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
341 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 | 342 | struct ceph_mds_request *req; |
73737682 | 343 | struct ceph_file_info *fi = file->private_data; |
124e68e7 SW |
344 | int err; |
345 | int flags, fmode, wanted; | |
346 | ||
73737682 | 347 | if (fi) { |
124e68e7 SW |
348 | dout("open file %p is already opened\n", file); |
349 | return 0; | |
350 | } | |
351 | ||
352 | /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */ | |
353 | flags = file->f_flags & ~(O_CREAT|O_EXCL); | |
354 | if (S_ISDIR(inode->i_mode)) | |
355 | flags = O_DIRECTORY; /* mds likes to know */ | |
356 | ||
357 | dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode, | |
358 | ceph_vinop(inode), file, flags, file->f_flags); | |
359 | fmode = ceph_flags_to_mode(flags); | |
360 | wanted = ceph_caps_for_mode(fmode); | |
361 | ||
362 | /* snapped files are read-only */ | |
363 | if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE)) | |
364 | return -EROFS; | |
365 | ||
366 | /* trivially open snapdir */ | |
367 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
be655596 | 368 | spin_lock(&ci->i_ceph_lock); |
124e68e7 | 369 | __ceph_get_fmode(ci, fmode); |
be655596 | 370 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
371 | return ceph_init_file(inode, file, fmode); |
372 | } | |
373 | ||
374 | /* | |
7421ab80 SW |
375 | * No need to block if we have caps on the auth MDS (for |
376 | * write) or any MDS (for read). Update wanted set | |
124e68e7 SW |
377 | * asynchronously. |
378 | */ | |
be655596 | 379 | spin_lock(&ci->i_ceph_lock); |
7421ab80 SW |
380 | if (__ceph_is_any_real_caps(ci) && |
381 | (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) { | |
c1944fed | 382 | int mds_wanted = __ceph_caps_mds_wanted(ci, true); |
124e68e7 SW |
383 | int issued = __ceph_caps_issued(ci, NULL); |
384 | ||
385 | dout("open %p fmode %d want %s issued %s using existing\n", | |
386 | inode, fmode, ceph_cap_string(wanted), | |
387 | ceph_cap_string(issued)); | |
388 | __ceph_get_fmode(ci, fmode); | |
be655596 | 389 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
390 | |
391 | /* adjust wanted? */ | |
392 | if ((issued & wanted) != wanted && | |
393 | (mds_wanted & wanted) != wanted && | |
394 | ceph_snap(inode) != CEPH_SNAPDIR) | |
395 | ceph_check_caps(ci, 0, NULL); | |
396 | ||
397 | return ceph_init_file(inode, file, fmode); | |
398 | } else if (ceph_snap(inode) != CEPH_NOSNAP && | |
399 | (ci->i_snap_caps & wanted) == wanted) { | |
400 | __ceph_get_fmode(ci, fmode); | |
be655596 | 401 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
402 | return ceph_init_file(inode, file, fmode); |
403 | } | |
99ccbd22 | 404 | |
be655596 | 405 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
406 | |
407 | dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted)); | |
408 | req = prepare_open_request(inode->i_sb, flags, 0); | |
409 | if (IS_ERR(req)) { | |
410 | err = PTR_ERR(req); | |
411 | goto out; | |
412 | } | |
70b666c3 SW |
413 | req->r_inode = inode; |
414 | ihold(inode); | |
99ccbd22 | 415 | |
124e68e7 | 416 | req->r_num_caps = 1; |
e36d571d | 417 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
124e68e7 SW |
418 | if (!err) |
419 | err = ceph_init_file(inode, file, req->r_fmode); | |
420 | ceph_mdsc_put_request(req); | |
421 | dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode)); | |
422 | out: | |
423 | return err; | |
424 | } | |
425 | ||
426 | ||
427 | /* | |
5ef50c3b SW |
428 | * Do a lookup + open with a single request. If we get a non-existent |
429 | * file or symlink, return 1 so the VFS can retry. | |
124e68e7 | 430 | */ |
5ef50c3b | 431 | int ceph_atomic_open(struct inode *dir, struct dentry *dentry, |
30d90494 | 432 | struct file *file, unsigned flags, umode_t mode, |
d9585277 | 433 | int *opened) |
124e68e7 | 434 | { |
3d14c5d2 YS |
435 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
436 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 | 437 | struct ceph_mds_request *req; |
5ef50c3b | 438 | struct dentry *dn; |
b1ee94aa | 439 | struct ceph_acls_info acls = {}; |
b7a29217 | 440 | int mask; |
124e68e7 | 441 | int err; |
124e68e7 | 442 | |
a455589f AV |
443 | dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n", |
444 | dir, dentry, dentry, | |
5ef50c3b SW |
445 | d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode); |
446 | ||
447 | if (dentry->d_name.len > NAME_MAX) | |
448 | return -ENAMETOOLONG; | |
449 | ||
b1ee94aa | 450 | if (flags & O_CREAT) { |
b7a29217 LH |
451 | if (ceph_quota_is_max_files_exceeded(dir)) |
452 | return -EDQUOT; | |
b1ee94aa YZ |
453 | err = ceph_pre_init_acls(dir, &mode, &acls); |
454 | if (err < 0) | |
455 | return err; | |
456 | } | |
457 | ||
124e68e7 SW |
458 | /* do the open */ |
459 | req = prepare_open_request(dir->i_sb, flags, mode); | |
b1ee94aa YZ |
460 | if (IS_ERR(req)) { |
461 | err = PTR_ERR(req); | |
462 | goto out_acl; | |
463 | } | |
124e68e7 SW |
464 | req->r_dentry = dget(dentry); |
465 | req->r_num_caps = 2; | |
466 | if (flags & O_CREAT) { | |
222b7f90 | 467 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL; |
124e68e7 | 468 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; |
b1ee94aa YZ |
469 | if (acls.pagelist) { |
470 | req->r_pagelist = acls.pagelist; | |
471 | acls.pagelist = NULL; | |
472 | } | |
124e68e7 | 473 | } |
315f2408 YZ |
474 | |
475 | mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED; | |
476 | if (ceph_security_xattr_wanted(dir)) | |
477 | mask |= CEPH_CAP_XATTR_SHARED; | |
478 | req->r_args.open.mask = cpu_to_le32(mask); | |
479 | ||
3dd69aab JL |
480 | req->r_parent = dir; |
481 | set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); | |
acda7657 SW |
482 | err = ceph_mdsc_do_request(mdsc, |
483 | (flags & (O_CREAT|O_TRUNC)) ? dir : NULL, | |
484 | req); | |
bf91c315 | 485 | err = ceph_handle_snapdir(req, dentry, err); |
79aec984 | 486 | if (err) |
b1ee94aa | 487 | goto out_req; |
79aec984 | 488 | |
a43137f7 | 489 | if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry) |
124e68e7 | 490 | err = ceph_handle_notrace_create(dir, dentry); |
2d83bde9 | 491 | |
00699ad8 | 492 | if (d_in_lookup(dentry)) { |
5ef50c3b SW |
493 | dn = ceph_finish_lookup(req, dentry, err); |
494 | if (IS_ERR(dn)) | |
495 | err = PTR_ERR(dn); | |
496 | } else { | |
497 | /* we were given a hashed negative dentry */ | |
498 | dn = NULL; | |
499 | } | |
500 | if (err) | |
b1ee94aa | 501 | goto out_req; |
2b0143b5 | 502 | if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) { |
5ef50c3b SW |
503 | /* make vfs retry on splice, ENOENT, or symlink */ |
504 | dout("atomic_open finish_no_open on dn %p\n", dn); | |
505 | err = finish_no_open(file, dn); | |
506 | } else { | |
507 | dout("atomic_open finish_open on dn %p\n", dn); | |
6e8575fa | 508 | if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) { |
2b0143b5 | 509 | ceph_init_inode_acls(d_inode(dentry), &acls); |
6e8575fa SL |
510 | *opened |= FILE_CREATED; |
511 | } | |
5ef50c3b SW |
512 | err = finish_open(file, dentry, ceph_open, opened); |
513 | } | |
b1ee94aa | 514 | out_req: |
ab866549 YZ |
515 | if (!req->r_err && req->r_target_inode) |
516 | ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode); | |
5ef50c3b | 517 | ceph_mdsc_put_request(req); |
b1ee94aa YZ |
518 | out_acl: |
519 | ceph_release_acls_info(&acls); | |
5ef50c3b | 520 | dout("atomic_open result=%d\n", err); |
d9585277 | 521 | return err; |
124e68e7 SW |
522 | } |
523 | ||
524 | int ceph_release(struct inode *inode, struct file *file) | |
525 | { | |
526 | struct ceph_inode_info *ci = ceph_inode(inode); | |
124e68e7 | 527 | |
bb48bd4d CX |
528 | if (S_ISDIR(inode->i_mode)) { |
529 | struct ceph_dir_file_info *dfi = file->private_data; | |
530 | dout("release inode %p dir file %p\n", inode, file); | |
531 | WARN_ON(!list_empty(&dfi->file_info.rw_contexts)); | |
532 | ||
533 | ceph_put_fmode(ci, dfi->file_info.fmode); | |
534 | ||
535 | if (dfi->last_readdir) | |
536 | ceph_mdsc_put_request(dfi->last_readdir); | |
537 | kfree(dfi->last_name); | |
538 | kfree(dfi->dir_info); | |
539 | kmem_cache_free(ceph_dir_file_cachep, dfi); | |
540 | } else { | |
541 | struct ceph_file_info *fi = file->private_data; | |
542 | dout("release inode %p regular file %p\n", inode, file); | |
543 | WARN_ON(!list_empty(&fi->rw_contexts)); | |
544 | ||
545 | ceph_put_fmode(ci, fi->fmode); | |
546 | kmem_cache_free(ceph_file_cachep, fi); | |
547 | } | |
195d3ce2 SW |
548 | |
549 | /* wake up anyone waiting for caps on this inode */ | |
03066f23 | 550 | wake_up_all(&ci->i_cap_wq); |
124e68e7 SW |
551 | return 0; |
552 | } | |
553 | ||
83701246 | 554 | enum { |
c8fe9b17 YZ |
555 | HAVE_RETRIED = 1, |
556 | CHECK_EOF = 2, | |
557 | READ_INLINE = 3, | |
83701246 YZ |
558 | }; |
559 | ||
124e68e7 SW |
560 | /* |
561 | * Read a range of bytes striped over one or more objects. Iterate over | |
562 | * objects we stripe over. (That's not atomic, but good enough for now.) | |
563 | * | |
564 | * If we get a short result from the OSD, check against i_size; we need to | |
565 | * only return a short read to the caller if we hit EOF. | |
566 | */ | |
567 | static int striped_read(struct inode *inode, | |
7ce469a5 | 568 | u64 pos, u64 len, |
6a026589 | 569 | struct page **pages, int num_pages, |
7ce469a5 | 570 | int page_align, int *checkeof) |
124e68e7 | 571 | { |
3d14c5d2 | 572 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
124e68e7 | 573 | struct ceph_inode_info *ci = ceph_inode(inode); |
7ce469a5 | 574 | u64 this_len; |
99c88e69 | 575 | loff_t i_size; |
7ce469a5 YZ |
576 | int page_idx; |
577 | int ret, read = 0; | |
124e68e7 SW |
578 | bool hit_stripe, was_short; |
579 | ||
580 | /* | |
581 | * we may need to do multiple reads. not atomic, unfortunately. | |
582 | */ | |
124e68e7 | 583 | more: |
7ce469a5 YZ |
584 | this_len = len; |
585 | page_idx = (page_align + read) >> PAGE_SHIFT; | |
3d14c5d2 | 586 | ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), |
124e68e7 | 587 | &ci->i_layout, pos, &this_len, |
7ce469a5 YZ |
588 | ci->i_truncate_seq, ci->i_truncate_size, |
589 | pages + page_idx, num_pages - page_idx, | |
590 | ((page_align + read) & ~PAGE_MASK)); | |
124e68e7 SW |
591 | if (ret == -ENOENT) |
592 | ret = 0; | |
7ce469a5 | 593 | hit_stripe = this_len < len; |
0e98728f | 594 | was_short = ret >= 0 && ret < this_len; |
7ce469a5 | 595 | dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, len, read, |
124e68e7 SW |
596 | ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : ""); |
597 | ||
99c88e69 | 598 | i_size = i_size_read(inode); |
02ae66d8 | 599 | if (ret >= 0) { |
99c88e69 YZ |
600 | if (was_short && (pos + ret < i_size)) { |
601 | int zlen = min(this_len - ret, i_size - pos - ret); | |
7ce469a5 | 602 | int zoff = page_align + read + ret; |
02ae66d8 | 603 | dout(" zero gap %llu to %llu\n", |
7ce469a5 | 604 | pos + ret, pos + ret + zlen); |
1487a688 YZ |
605 | ceph_zero_page_vector_range(zoff, zlen, pages); |
606 | ret += zlen; | |
124e68e7 | 607 | } |
02ae66d8 | 608 | |
7ce469a5 | 609 | read += ret; |
124e68e7 | 610 | pos += ret; |
7ce469a5 | 611 | len -= ret; |
124e68e7 | 612 | |
02ae66d8 | 613 | /* hit stripe and need continue*/ |
7ce469a5 | 614 | if (len && hit_stripe && pos < i_size) |
124e68e7 SW |
615 | goto more; |
616 | } | |
617 | ||
ee7289bf | 618 | if (read > 0) { |
02ae66d8 | 619 | ret = read; |
c3cd6283 | 620 | /* did we bounce off eof? */ |
7ce469a5 | 621 | if (pos + len > i_size) |
83701246 | 622 | *checkeof = CHECK_EOF; |
124e68e7 SW |
623 | } |
624 | ||
124e68e7 SW |
625 | dout("striped_read returns %d\n", ret); |
626 | return ret; | |
627 | } | |
628 | ||
629 | /* | |
630 | * Completely synchronous read and write methods. Direct from __user | |
631 | * buffer to osd, or directly to user pages (if O_DIRECT). | |
632 | * | |
633 | * If the read spans object boundary, just do multiple reads. | |
634 | */ | |
7ce469a5 YZ |
635 | static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to, |
636 | int *checkeof) | |
124e68e7 | 637 | { |
8eb4efb0 | 638 | struct file *file = iocb->ki_filp; |
496ad9aa | 639 | struct inode *inode = file_inode(file); |
124e68e7 | 640 | struct page **pages; |
8eb4efb0 | 641 | u64 off = iocb->ki_pos; |
7ce469a5 YZ |
642 | int num_pages; |
643 | ssize_t ret; | |
644 | size_t len = iov_iter_count(to); | |
124e68e7 | 645 | |
1c0a9c2d | 646 | dout("sync_read on file %p %llu~%u %s\n", file, off, (unsigned)len, |
124e68e7 | 647 | (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); |
d0d0db22 YZ |
648 | |
649 | if (!len) | |
650 | return 0; | |
e98b6fed SW |
651 | /* |
652 | * flush any page cache pages in this range. this | |
653 | * will make concurrent normal and sync io slow, | |
654 | * but it will at least behave sensibly when they are | |
655 | * in sequence. | |
656 | */ | |
8eb4efb0 | 657 | ret = filemap_write_and_wait_range(inode->i_mapping, off, |
658 | off + len); | |
29065a51 | 659 | if (ret < 0) |
8eb4efb0 | 660 | return ret; |
29065a51 | 661 | |
7ce469a5 YZ |
662 | if (unlikely(to->type & ITER_PIPE)) { |
663 | size_t page_off; | |
664 | ret = iov_iter_get_pages_alloc(to, &pages, len, | |
665 | &page_off); | |
666 | if (ret <= 0) | |
667 | return -ENOMEM; | |
668 | num_pages = DIV_ROUND_UP(ret + page_off, PAGE_SIZE); | |
669 | ||
670 | ret = striped_read(inode, off, ret, pages, num_pages, | |
671 | page_off, checkeof); | |
672 | if (ret > 0) { | |
673 | iov_iter_advance(to, ret); | |
674 | off += ret; | |
675 | } else { | |
676 | iov_iter_advance(to, 0); | |
677 | } | |
678 | ceph_put_page_vector(pages, num_pages, false); | |
679 | } else { | |
680 | num_pages = calc_pages_for(off, len); | |
681 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); | |
682 | if (IS_ERR(pages)) | |
683 | return PTR_ERR(pages); | |
684 | ||
685 | ret = striped_read(inode, off, len, pages, num_pages, | |
686 | (off & ~PAGE_MASK), checkeof); | |
687 | if (ret > 0) { | |
688 | int l, k = 0; | |
689 | size_t left = ret; | |
690 | ||
691 | while (left) { | |
692 | size_t page_off = off & ~PAGE_MASK; | |
693 | size_t copy = min_t(size_t, left, | |
694 | PAGE_SIZE - page_off); | |
695 | l = copy_page_to_iter(pages[k++], page_off, | |
696 | copy, to); | |
697 | off += l; | |
698 | left -= l; | |
699 | if (l < copy) | |
700 | break; | |
701 | } | |
8eb4efb0 | 702 | } |
7ce469a5 | 703 | ceph_release_page_vector(pages, num_pages); |
8eb4efb0 | 704 | } |
124e68e7 | 705 | |
8eb4efb0 | 706 | if (off > iocb->ki_pos) { |
707 | ret = off - iocb->ki_pos; | |
708 | iocb->ki_pos = off; | |
709 | } | |
124e68e7 | 710 | |
7ce469a5 | 711 | dout("sync_read result %zd\n", ret); |
124e68e7 SW |
712 | return ret; |
713 | } | |
714 | ||
c8fe9b17 YZ |
715 | struct ceph_aio_request { |
716 | struct kiocb *iocb; | |
717 | size_t total_len; | |
85784f93 YZ |
718 | bool write; |
719 | bool should_dirty; | |
c8fe9b17 YZ |
720 | int error; |
721 | struct list_head osd_reqs; | |
722 | unsigned num_reqs; | |
723 | atomic_t pending_reqs; | |
5be0389d | 724 | struct timespec mtime; |
c8fe9b17 YZ |
725 | struct ceph_cap_flush *prealloc_cf; |
726 | }; | |
727 | ||
5be0389d YZ |
728 | struct ceph_aio_work { |
729 | struct work_struct work; | |
730 | struct ceph_osd_request *req; | |
731 | }; | |
732 | ||
733 | static void ceph_aio_retry_work(struct work_struct *work); | |
734 | ||
c8fe9b17 YZ |
735 | static void ceph_aio_complete(struct inode *inode, |
736 | struct ceph_aio_request *aio_req) | |
737 | { | |
738 | struct ceph_inode_info *ci = ceph_inode(inode); | |
739 | int ret; | |
740 | ||
741 | if (!atomic_dec_and_test(&aio_req->pending_reqs)) | |
742 | return; | |
743 | ||
744 | ret = aio_req->error; | |
745 | if (!ret) | |
746 | ret = aio_req->total_len; | |
747 | ||
748 | dout("ceph_aio_complete %p rc %d\n", inode, ret); | |
749 | ||
750 | if (ret >= 0 && aio_req->write) { | |
751 | int dirty; | |
752 | ||
753 | loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len; | |
754 | if (endoff > i_size_read(inode)) { | |
755 | if (ceph_inode_set_size(inode, endoff)) | |
756 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); | |
757 | } | |
758 | ||
759 | spin_lock(&ci->i_ceph_lock); | |
760 | ci->i_inline_version = CEPH_INLINE_NONE; | |
761 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, | |
762 | &aio_req->prealloc_cf); | |
763 | spin_unlock(&ci->i_ceph_lock); | |
764 | if (dirty) | |
765 | __mark_inode_dirty(inode, dirty); | |
766 | ||
767 | } | |
768 | ||
769 | ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR : | |
770 | CEPH_CAP_FILE_RD)); | |
771 | ||
772 | aio_req->iocb->ki_complete(aio_req->iocb, ret, 0); | |
773 | ||
774 | ceph_free_cap_flush(aio_req->prealloc_cf); | |
775 | kfree(aio_req); | |
776 | } | |
777 | ||
85e084fe | 778 | static void ceph_aio_complete_req(struct ceph_osd_request *req) |
c8fe9b17 YZ |
779 | { |
780 | int rc = req->r_result; | |
781 | struct inode *inode = req->r_inode; | |
782 | struct ceph_aio_request *aio_req = req->r_priv; | |
783 | struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0); | |
c8fe9b17 | 784 | |
fc218544 ID |
785 | BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_BVECS); |
786 | BUG_ON(!osd_data->num_bvecs); | |
787 | ||
788 | dout("ceph_aio_complete_req %p rc %d bytes %u\n", | |
789 | inode, rc, osd_data->bvec_pos.iter.bi_size); | |
c8fe9b17 YZ |
790 | |
791 | if (rc == -EOLDSNAPC) { | |
5be0389d YZ |
792 | struct ceph_aio_work *aio_work; |
793 | BUG_ON(!aio_req->write); | |
794 | ||
795 | aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS); | |
796 | if (aio_work) { | |
797 | INIT_WORK(&aio_work->work, ceph_aio_retry_work); | |
798 | aio_work->req = req; | |
799 | queue_work(ceph_inode_to_client(inode)->wb_wq, | |
800 | &aio_work->work); | |
801 | return; | |
802 | } | |
803 | rc = -ENOMEM; | |
804 | } else if (!aio_req->write) { | |
c8fe9b17 YZ |
805 | if (rc == -ENOENT) |
806 | rc = 0; | |
fc218544 ID |
807 | if (rc >= 0 && osd_data->bvec_pos.iter.bi_size > rc) { |
808 | struct iov_iter i; | |
809 | int zlen = osd_data->bvec_pos.iter.bi_size - rc; | |
810 | ||
c8fe9b17 YZ |
811 | /* |
812 | * If read is satisfied by single OSD request, | |
813 | * it can pass EOF. Otherwise read is within | |
814 | * i_size. | |
815 | */ | |
816 | if (aio_req->num_reqs == 1) { | |
817 | loff_t i_size = i_size_read(inode); | |
818 | loff_t endoff = aio_req->iocb->ki_pos + rc; | |
819 | if (endoff < i_size) | |
820 | zlen = min_t(size_t, zlen, | |
821 | i_size - endoff); | |
822 | aio_req->total_len = rc + zlen; | |
823 | } | |
824 | ||
fc218544 ID |
825 | iov_iter_bvec(&i, ITER_BVEC, osd_data->bvec_pos.bvecs, |
826 | osd_data->num_bvecs, | |
827 | osd_data->bvec_pos.iter.bi_size); | |
828 | iov_iter_advance(&i, rc); | |
829 | iov_iter_zero(zlen, &i); | |
c8fe9b17 YZ |
830 | } |
831 | } | |
832 | ||
fc218544 ID |
833 | put_bvecs(osd_data->bvec_pos.bvecs, osd_data->num_bvecs, |
834 | aio_req->should_dirty); | |
c8fe9b17 YZ |
835 | ceph_osdc_put_request(req); |
836 | ||
837 | if (rc < 0) | |
838 | cmpxchg(&aio_req->error, 0, rc); | |
839 | ||
840 | ceph_aio_complete(inode, aio_req); | |
841 | return; | |
842 | } | |
843 | ||
5be0389d YZ |
844 | static void ceph_aio_retry_work(struct work_struct *work) |
845 | { | |
846 | struct ceph_aio_work *aio_work = | |
847 | container_of(work, struct ceph_aio_work, work); | |
848 | struct ceph_osd_request *orig_req = aio_work->req; | |
849 | struct ceph_aio_request *aio_req = orig_req->r_priv; | |
850 | struct inode *inode = orig_req->r_inode; | |
851 | struct ceph_inode_info *ci = ceph_inode(inode); | |
852 | struct ceph_snap_context *snapc; | |
853 | struct ceph_osd_request *req; | |
854 | int ret; | |
855 | ||
856 | spin_lock(&ci->i_ceph_lock); | |
857 | if (__ceph_have_pending_cap_snap(ci)) { | |
858 | struct ceph_cap_snap *capsnap = | |
859 | list_last_entry(&ci->i_cap_snaps, | |
860 | struct ceph_cap_snap, | |
861 | ci_item); | |
862 | snapc = ceph_get_snap_context(capsnap->context); | |
863 | } else { | |
864 | BUG_ON(!ci->i_head_snapc); | |
865 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
866 | } | |
867 | spin_unlock(&ci->i_ceph_lock); | |
868 | ||
869 | req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 2, | |
870 | false, GFP_NOFS); | |
1418bf07 DC |
871 | if (!req) { |
872 | ret = -ENOMEM; | |
5be0389d YZ |
873 | req = orig_req; |
874 | goto out; | |
875 | } | |
876 | ||
b178cf43 | 877 | req->r_flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE; |
63244fa1 | 878 | ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc); |
d30291b9 | 879 | ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid); |
5be0389d | 880 | |
13d1ad16 ID |
881 | ret = ceph_osdc_alloc_messages(req, GFP_NOFS); |
882 | if (ret) { | |
883 | ceph_osdc_put_request(req); | |
884 | req = orig_req; | |
885 | goto out; | |
886 | } | |
5be0389d YZ |
887 | |
888 | req->r_ops[0] = orig_req->r_ops[0]; | |
5be0389d | 889 | |
bb873b53 ID |
890 | req->r_mtime = aio_req->mtime; |
891 | req->r_data_offset = req->r_ops[0].extent.offset; | |
5be0389d | 892 | |
5be0389d YZ |
893 | ceph_osdc_put_request(orig_req); |
894 | ||
895 | req->r_callback = ceph_aio_complete_req; | |
896 | req->r_inode = inode; | |
897 | req->r_priv = aio_req; | |
a1f4020a | 898 | req->r_abort_on_full = true; |
5be0389d YZ |
899 | |
900 | ret = ceph_osdc_start_request(req->r_osdc, req, false); | |
901 | out: | |
902 | if (ret < 0) { | |
5be0389d | 903 | req->r_result = ret; |
85e084fe | 904 | ceph_aio_complete_req(req); |
5be0389d YZ |
905 | } |
906 | ||
db6aed70 | 907 | ceph_put_snap_context(snapc); |
5be0389d YZ |
908 | kfree(aio_work); |
909 | } | |
910 | ||
e8344e66 | 911 | static ssize_t |
c8fe9b17 YZ |
912 | ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter, |
913 | struct ceph_snap_context *snapc, | |
914 | struct ceph_cap_flush **pcf) | |
124e68e7 | 915 | { |
e8344e66 | 916 | struct file *file = iocb->ki_filp; |
496ad9aa | 917 | struct inode *inode = file_inode(file); |
124e68e7 | 918 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 | 919 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
acead002 | 920 | struct ceph_vino vino; |
124e68e7 | 921 | struct ceph_osd_request *req; |
fc218544 | 922 | struct bio_vec *bvecs; |
c8fe9b17 YZ |
923 | struct ceph_aio_request *aio_req = NULL; |
924 | int num_pages = 0; | |
124e68e7 | 925 | int flags; |
124e68e7 | 926 | int ret; |
c2050a45 | 927 | struct timespec mtime = current_time(inode); |
c8fe9b17 YZ |
928 | size_t count = iov_iter_count(iter); |
929 | loff_t pos = iocb->ki_pos; | |
930 | bool write = iov_iter_rw(iter) == WRITE; | |
85784f93 | 931 | bool should_dirty = !write && iter_is_iovec(iter); |
124e68e7 | 932 | |
c8fe9b17 | 933 | if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
124e68e7 SW |
934 | return -EROFS; |
935 | ||
1c0a9c2d YZ |
936 | dout("sync_direct_%s on file %p %lld~%u snapc %p seq %lld\n", |
937 | (write ? "write" : "read"), file, pos, (unsigned)count, | |
938 | snapc, snapc->seq); | |
124e68e7 | 939 | |
e8344e66 | 940 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); |
29065a51 YS |
941 | if (ret < 0) |
942 | return ret; | |
943 | ||
c8fe9b17 | 944 | if (write) { |
5d7eb1a3 | 945 | int ret2 = invalidate_inode_pages2_range(inode->i_mapping, |
09cbfeaf KS |
946 | pos >> PAGE_SHIFT, |
947 | (pos + count) >> PAGE_SHIFT); | |
5d7eb1a3 | 948 | if (ret2 < 0) |
a380a031 | 949 | dout("invalidate_inode_pages2_range returned %d\n", ret2); |
29065a51 | 950 | |
b178cf43 | 951 | flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE; |
c8fe9b17 YZ |
952 | } else { |
953 | flags = CEPH_OSD_FLAG_READ; | |
954 | } | |
124e68e7 | 955 | |
c8fe9b17 | 956 | while (iov_iter_count(iter) > 0) { |
fc218544 | 957 | u64 size = iov_iter_count(iter); |
c8fe9b17 | 958 | ssize_t len; |
e8344e66 | 959 | |
3a15b38f ID |
960 | if (write) |
961 | size = min_t(u64, size, fsc->mount_options->wsize); | |
962 | else | |
963 | size = min_t(u64, size, fsc->mount_options->rsize); | |
964 | ||
e8344e66 | 965 | vino = ceph_vino(inode); |
966 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
c8fe9b17 | 967 | vino, pos, &size, 0, |
3fb99d48 | 968 | 1, |
c8fe9b17 YZ |
969 | write ? CEPH_OSD_OP_WRITE : |
970 | CEPH_OSD_OP_READ, | |
971 | flags, snapc, | |
e8344e66 | 972 | ci->i_truncate_seq, |
973 | ci->i_truncate_size, | |
974 | false); | |
975 | if (IS_ERR(req)) { | |
976 | ret = PTR_ERR(req); | |
eab87235 | 977 | break; |
e8344e66 | 978 | } |
124e68e7 | 979 | |
fc218544 ID |
980 | len = iter_get_bvecs_alloc(iter, size, &bvecs, &num_pages); |
981 | if (len < 0) { | |
64c31311 | 982 | ceph_osdc_put_request(req); |
fc218544 | 983 | ret = len; |
64c31311 | 984 | break; |
124e68e7 | 985 | } |
fc218544 ID |
986 | if (len != size) |
987 | osd_req_op_extent_update(req, 0, len); | |
124e68e7 SW |
988 | |
989 | /* | |
c8fe9b17 YZ |
990 | * To simplify error handling, allow AIO when IO within i_size |
991 | * or IO can be satisfied by single OSD request. | |
124e68e7 | 992 | */ |
c8fe9b17 YZ |
993 | if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) && |
994 | (len == count || pos + count <= i_size_read(inode))) { | |
995 | aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL); | |
996 | if (aio_req) { | |
997 | aio_req->iocb = iocb; | |
998 | aio_req->write = write; | |
85784f93 | 999 | aio_req->should_dirty = should_dirty; |
c8fe9b17 YZ |
1000 | INIT_LIST_HEAD(&aio_req->osd_reqs); |
1001 | if (write) { | |
5be0389d | 1002 | aio_req->mtime = mtime; |
c8fe9b17 YZ |
1003 | swap(aio_req->prealloc_cf, *pcf); |
1004 | } | |
1005 | } | |
1006 | /* ignore error */ | |
1007 | } | |
1008 | ||
1009 | if (write) { | |
1010 | /* | |
1011 | * throw out any page cache pages in this range. this | |
1012 | * may block. | |
1013 | */ | |
1014 | truncate_inode_pages_range(inode->i_mapping, pos, | |
09cbfeaf | 1015 | (pos+len) | (PAGE_SIZE - 1)); |
c8fe9b17 | 1016 | |
bb873b53 | 1017 | req->r_mtime = mtime; |
c8fe9b17 YZ |
1018 | } |
1019 | ||
fc218544 | 1020 | osd_req_op_extent_osd_data_bvecs(req, 0, bvecs, num_pages, len); |
e8344e66 | 1021 | |
c8fe9b17 YZ |
1022 | if (aio_req) { |
1023 | aio_req->total_len += len; | |
1024 | aio_req->num_reqs++; | |
1025 | atomic_inc(&aio_req->pending_reqs); | |
1026 | ||
1027 | req->r_callback = ceph_aio_complete_req; | |
1028 | req->r_inode = inode; | |
1029 | req->r_priv = aio_req; | |
1030 | list_add_tail(&req->r_unsafe_item, &aio_req->osd_reqs); | |
1031 | ||
1032 | pos += len; | |
c8fe9b17 YZ |
1033 | continue; |
1034 | } | |
1035 | ||
1036 | ret = ceph_osdc_start_request(req->r_osdc, req, false); | |
e8344e66 | 1037 | if (!ret) |
1038 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1039 | ||
c8fe9b17 YZ |
1040 | size = i_size_read(inode); |
1041 | if (!write) { | |
1042 | if (ret == -ENOENT) | |
1043 | ret = 0; | |
1044 | if (ret >= 0 && ret < len && pos + ret < size) { | |
fc218544 | 1045 | struct iov_iter i; |
c8fe9b17 YZ |
1046 | int zlen = min_t(size_t, len - ret, |
1047 | size - pos - ret); | |
fc218544 ID |
1048 | |
1049 | iov_iter_bvec(&i, ITER_BVEC, bvecs, num_pages, | |
1050 | len); | |
1051 | iov_iter_advance(&i, ret); | |
1052 | iov_iter_zero(zlen, &i); | |
c8fe9b17 YZ |
1053 | ret += zlen; |
1054 | } | |
1055 | if (ret >= 0) | |
1056 | len = ret; | |
1057 | } | |
1058 | ||
fc218544 | 1059 | put_bvecs(bvecs, num_pages, should_dirty); |
e8344e66 | 1060 | ceph_osdc_put_request(req); |
c8fe9b17 | 1061 | if (ret < 0) |
e8344e66 | 1062 | break; |
64c31311 | 1063 | |
c8fe9b17 | 1064 | pos += len; |
c8fe9b17 | 1065 | if (!write && pos >= size) |
e8344e66 | 1066 | break; |
64c31311 | 1067 | |
c8fe9b17 YZ |
1068 | if (write && pos > size) { |
1069 | if (ceph_inode_set_size(inode, pos)) | |
64c31311 AV |
1070 | ceph_check_caps(ceph_inode(inode), |
1071 | CHECK_CAPS_AUTHONLY, | |
1072 | NULL); | |
1073 | } | |
e8344e66 | 1074 | } |
1075 | ||
c8fe9b17 | 1076 | if (aio_req) { |
fc8c3892 YZ |
1077 | LIST_HEAD(osd_reqs); |
1078 | ||
c8fe9b17 YZ |
1079 | if (aio_req->num_reqs == 0) { |
1080 | kfree(aio_req); | |
1081 | return ret; | |
1082 | } | |
1083 | ||
1084 | ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR : | |
1085 | CEPH_CAP_FILE_RD); | |
1086 | ||
fc8c3892 YZ |
1087 | list_splice(&aio_req->osd_reqs, &osd_reqs); |
1088 | while (!list_empty(&osd_reqs)) { | |
1089 | req = list_first_entry(&osd_reqs, | |
c8fe9b17 YZ |
1090 | struct ceph_osd_request, |
1091 | r_unsafe_item); | |
1092 | list_del_init(&req->r_unsafe_item); | |
1093 | if (ret >= 0) | |
1094 | ret = ceph_osdc_start_request(req->r_osdc, | |
1095 | req, false); | |
1096 | if (ret < 0) { | |
1097 | req->r_result = ret; | |
85e084fe | 1098 | ceph_aio_complete_req(req); |
c8fe9b17 YZ |
1099 | } |
1100 | } | |
1101 | return -EIOCBQUEUED; | |
1102 | } | |
1103 | ||
1104 | if (ret != -EOLDSNAPC && pos > iocb->ki_pos) { | |
1105 | ret = pos - iocb->ki_pos; | |
e8344e66 | 1106 | iocb->ki_pos = pos; |
e8344e66 | 1107 | } |
1108 | return ret; | |
1109 | } | |
1110 | ||
e8344e66 | 1111 | /* |
1112 | * Synchronous write, straight from __user pointer or user pages. | |
1113 | * | |
1114 | * If write spans object boundary, just do multiple writes. (For a | |
1115 | * correct atomic write, we should e.g. take write locks on all | |
1116 | * objects, rollback on failure, etc.) | |
1117 | */ | |
06fee30f | 1118 | static ssize_t |
5dda377c YZ |
1119 | ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos, |
1120 | struct ceph_snap_context *snapc) | |
e8344e66 | 1121 | { |
1122 | struct file *file = iocb->ki_filp; | |
1123 | struct inode *inode = file_inode(file); | |
1124 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1125 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
e8344e66 | 1126 | struct ceph_vino vino; |
1127 | struct ceph_osd_request *req; | |
1128 | struct page **pages; | |
1129 | u64 len; | |
1130 | int num_pages; | |
1131 | int written = 0; | |
1132 | int flags; | |
e8344e66 | 1133 | int ret; |
efb0ca76 | 1134 | bool check_caps = false; |
c2050a45 | 1135 | struct timespec mtime = current_time(inode); |
4908b822 | 1136 | size_t count = iov_iter_count(from); |
e8344e66 | 1137 | |
1138 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) | |
1139 | return -EROFS; | |
1140 | ||
1c0a9c2d YZ |
1141 | dout("sync_write on file %p %lld~%u snapc %p seq %lld\n", |
1142 | file, pos, (unsigned)count, snapc, snapc->seq); | |
e8344e66 | 1143 | |
1144 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); | |
1145 | if (ret < 0) | |
1146 | return ret; | |
1147 | ||
1148 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
09cbfeaf KS |
1149 | pos >> PAGE_SHIFT, |
1150 | (pos + count) >> PAGE_SHIFT); | |
e8344e66 | 1151 | if (ret < 0) |
1152 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
1153 | ||
b178cf43 | 1154 | flags = /* CEPH_OSD_FLAG_ORDERSNAP | */ CEPH_OSD_FLAG_WRITE; |
e8344e66 | 1155 | |
4908b822 | 1156 | while ((len = iov_iter_count(from)) > 0) { |
e8344e66 | 1157 | size_t left; |
1158 | int n; | |
1159 | ||
e8344e66 | 1160 | vino = ceph_vino(inode); |
1161 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
715e4cd4 | 1162 | vino, pos, &len, 0, 1, |
e8344e66 | 1163 | CEPH_OSD_OP_WRITE, flags, snapc, |
1164 | ci->i_truncate_seq, | |
1165 | ci->i_truncate_size, | |
1166 | false); | |
1167 | if (IS_ERR(req)) { | |
1168 | ret = PTR_ERR(req); | |
eab87235 | 1169 | break; |
e8344e66 | 1170 | } |
1171 | ||
1172 | /* | |
1173 | * write from beginning of first page, | |
1174 | * regardless of io alignment | |
1175 | */ | |
09cbfeaf | 1176 | num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
e8344e66 | 1177 | |
687265e5 | 1178 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
124e68e7 SW |
1179 | if (IS_ERR(pages)) { |
1180 | ret = PTR_ERR(pages); | |
1181 | goto out; | |
1182 | } | |
e8344e66 | 1183 | |
1184 | left = len; | |
1185 | for (n = 0; n < num_pages; n++) { | |
125d725c | 1186 | size_t plen = min_t(size_t, left, PAGE_SIZE); |
4908b822 | 1187 | ret = copy_page_from_iter(pages[n], 0, plen, from); |
e8344e66 | 1188 | if (ret != plen) { |
1189 | ret = -EFAULT; | |
1190 | break; | |
1191 | } | |
1192 | left -= ret; | |
e8344e66 | 1193 | } |
1194 | ||
124e68e7 SW |
1195 | if (ret < 0) { |
1196 | ceph_release_page_vector(pages, num_pages); | |
1197 | goto out; | |
1198 | } | |
1199 | ||
e8344e66 | 1200 | req->r_inode = inode; |
124e68e7 | 1201 | |
e8344e66 | 1202 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, |
1203 | false, true); | |
02ee07d3 | 1204 | |
bb873b53 | 1205 | req->r_mtime = mtime; |
e8344e66 | 1206 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
1207 | if (!ret) | |
1208 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
124e68e7 SW |
1209 | |
1210 | out: | |
e8344e66 | 1211 | ceph_osdc_put_request(req); |
26544c62 JL |
1212 | if (ret != 0) { |
1213 | ceph_set_error_write(ci); | |
e8344e66 | 1214 | break; |
26544c62 JL |
1215 | } |
1216 | ||
1217 | ceph_clear_error_write(ci); | |
1218 | pos += len; | |
1219 | written += len; | |
1220 | if (pos > i_size_read(inode)) { | |
1221 | check_caps = ceph_inode_set_size(inode, pos); | |
1222 | if (check_caps) | |
1223 | ceph_check_caps(ceph_inode(inode), | |
1224 | CHECK_CAPS_AUTHONLY, | |
1225 | NULL); | |
1226 | } | |
1227 | ||
e8344e66 | 1228 | } |
124e68e7 | 1229 | |
e8344e66 | 1230 | if (ret != -EOLDSNAPC && written > 0) { |
124e68e7 | 1231 | ret = written; |
e8344e66 | 1232 | iocb->ki_pos = pos; |
124e68e7 SW |
1233 | } |
1234 | return ret; | |
1235 | } | |
1236 | ||
1237 | /* | |
1238 | * Wrap generic_file_aio_read with checks for cap bits on the inode. | |
1239 | * Atomically grab references, so that those bits are not released | |
1240 | * back to the MDS mid-read. | |
1241 | * | |
1242 | * Hmm, the sync read case isn't actually async... should it be? | |
1243 | */ | |
3644424d | 1244 | static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) |
124e68e7 SW |
1245 | { |
1246 | struct file *filp = iocb->ki_filp; | |
2962507c | 1247 | struct ceph_file_info *fi = filp->private_data; |
66ee59af | 1248 | size_t len = iov_iter_count(to); |
496ad9aa | 1249 | struct inode *inode = file_inode(filp); |
124e68e7 | 1250 | struct ceph_inode_info *ci = ceph_inode(inode); |
3738daa6 | 1251 | struct page *pinned_page = NULL; |
124e68e7 | 1252 | ssize_t ret; |
2962507c | 1253 | int want, got = 0; |
83701246 | 1254 | int retry_op = 0, read = 0; |
124e68e7 | 1255 | |
6a026589 | 1256 | again: |
8eb4efb0 | 1257 | dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n", |
1258 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode); | |
1259 | ||
2962507c SW |
1260 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
1261 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
1262 | else | |
1263 | want = CEPH_CAP_FILE_CACHE; | |
3738daa6 | 1264 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page); |
124e68e7 | 1265 | if (ret < 0) |
8eb4efb0 | 1266 | return ret; |
124e68e7 | 1267 | |
2962507c | 1268 | if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 || |
2ba48ce5 | 1269 | (iocb->ki_flags & IOCB_DIRECT) || |
8eb4efb0 | 1270 | (fi->flags & CEPH_F_SYNC)) { |
8eb4efb0 | 1271 | |
1272 | dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n", | |
1273 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, | |
1274 | ceph_cap_string(got)); | |
1275 | ||
83701246 | 1276 | if (ci->i_inline_version == CEPH_INLINE_NONE) { |
c8fe9b17 YZ |
1277 | if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) { |
1278 | ret = ceph_direct_read_write(iocb, to, | |
1279 | NULL, NULL); | |
1280 | if (ret >= 0 && ret < len) | |
1281 | retry_op = CHECK_EOF; | |
1282 | } else { | |
1283 | ret = ceph_sync_read(iocb, to, &retry_op); | |
1284 | } | |
83701246 YZ |
1285 | } else { |
1286 | retry_op = READ_INLINE; | |
1287 | } | |
8eb4efb0 | 1288 | } else { |
5d988308 | 1289 | CEPH_DEFINE_RW_CONTEXT(rw_ctx, got); |
8eb4efb0 | 1290 | dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n", |
3644424d | 1291 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, |
8eb4efb0 | 1292 | ceph_cap_string(got)); |
5d988308 | 1293 | ceph_add_rw_context(fi, &rw_ctx); |
3644424d | 1294 | ret = generic_file_read_iter(iocb, to); |
5d988308 | 1295 | ceph_del_rw_context(fi, &rw_ctx); |
8eb4efb0 | 1296 | } |
124e68e7 SW |
1297 | dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", |
1298 | inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); | |
3738daa6 | 1299 | if (pinned_page) { |
09cbfeaf | 1300 | put_page(pinned_page); |
3738daa6 YZ |
1301 | pinned_page = NULL; |
1302 | } | |
124e68e7 | 1303 | ceph_put_cap_refs(ci, got); |
c8fe9b17 | 1304 | if (retry_op > HAVE_RETRIED && ret >= 0) { |
83701246 YZ |
1305 | int statret; |
1306 | struct page *page = NULL; | |
1307 | loff_t i_size; | |
1308 | if (retry_op == READ_INLINE) { | |
687265e5 | 1309 | page = __page_cache_alloc(GFP_KERNEL); |
83701246 YZ |
1310 | if (!page) |
1311 | return -ENOMEM; | |
1312 | } | |
6a026589 | 1313 | |
83701246 YZ |
1314 | statret = __ceph_do_getattr(inode, page, |
1315 | CEPH_STAT_CAP_INLINE_DATA, !!page); | |
1316 | if (statret < 0) { | |
0d7718f6 NB |
1317 | if (page) |
1318 | __free_page(page); | |
83701246 YZ |
1319 | if (statret == -ENODATA) { |
1320 | BUG_ON(retry_op != READ_INLINE); | |
1321 | goto again; | |
1322 | } | |
1323 | return statret; | |
1324 | } | |
6a026589 | 1325 | |
83701246 YZ |
1326 | i_size = i_size_read(inode); |
1327 | if (retry_op == READ_INLINE) { | |
fcc02d2a YZ |
1328 | BUG_ON(ret > 0 || read > 0); |
1329 | if (iocb->ki_pos < i_size && | |
09cbfeaf | 1330 | iocb->ki_pos < PAGE_SIZE) { |
83701246 YZ |
1331 | loff_t end = min_t(loff_t, i_size, |
1332 | iocb->ki_pos + len); | |
09cbfeaf | 1333 | end = min_t(loff_t, end, PAGE_SIZE); |
83701246 YZ |
1334 | if (statret < end) |
1335 | zero_user_segment(page, statret, end); | |
1336 | ret = copy_page_to_iter(page, | |
1337 | iocb->ki_pos & ~PAGE_MASK, | |
1338 | end - iocb->ki_pos, to); | |
1339 | iocb->ki_pos += ret; | |
fcc02d2a YZ |
1340 | read += ret; |
1341 | } | |
1342 | if (iocb->ki_pos < i_size && read < len) { | |
1343 | size_t zlen = min_t(size_t, len - read, | |
1344 | i_size - iocb->ki_pos); | |
1345 | ret = iov_iter_zero(zlen, to); | |
1346 | iocb->ki_pos += ret; | |
1347 | read += ret; | |
83701246 YZ |
1348 | } |
1349 | __free_pages(page, 0); | |
fcc02d2a | 1350 | return read; |
83701246 | 1351 | } |
6a026589 SW |
1352 | |
1353 | /* hit EOF or hole? */ | |
83701246 | 1354 | if (retry_op == CHECK_EOF && iocb->ki_pos < i_size && |
fcc02d2a | 1355 | ret < len) { |
8eb4efb0 | 1356 | dout("sync_read hit hole, ppos %lld < size %lld" |
99c88e69 | 1357 | ", reading more\n", iocb->ki_pos, i_size); |
8eb4efb0 | 1358 | |
6a026589 | 1359 | read += ret; |
6a026589 | 1360 | len -= ret; |
c8fe9b17 | 1361 | retry_op = HAVE_RETRIED; |
6a026589 SW |
1362 | goto again; |
1363 | } | |
1364 | } | |
8eb4efb0 | 1365 | |
6a026589 SW |
1366 | if (ret >= 0) |
1367 | ret += read; | |
1368 | ||
124e68e7 SW |
1369 | return ret; |
1370 | } | |
1371 | ||
1372 | /* | |
1373 | * Take cap references to avoid releasing caps to MDS mid-write. | |
1374 | * | |
1375 | * If we are synchronous, and write with an old snap context, the OSD | |
1376 | * may return EOLDSNAPC. In that case, retry the write.. _after_ | |
1377 | * dropping our cap refs and allowing the pending snap to logically | |
1378 | * complete _before_ this write occurs. | |
1379 | * | |
1380 | * If we are near ENOSPC, write synchronously. | |
1381 | */ | |
4908b822 | 1382 | static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from) |
124e68e7 SW |
1383 | { |
1384 | struct file *file = iocb->ki_filp; | |
33caad32 | 1385 | struct ceph_file_info *fi = file->private_data; |
496ad9aa | 1386 | struct inode *inode = file_inode(file); |
124e68e7 | 1387 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 YS |
1388 | struct ceph_osd_client *osdc = |
1389 | &ceph_sb_to_client(inode->i_sb)->client->osdc; | |
f66fd9f0 | 1390 | struct ceph_cap_flush *prealloc_cf; |
3309dd04 | 1391 | ssize_t count, written = 0; |
03d254ed | 1392 | int err, want, got; |
3309dd04 | 1393 | loff_t pos; |
124e68e7 SW |
1394 | |
1395 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
1396 | return -EROFS; | |
1397 | ||
f66fd9f0 YZ |
1398 | prealloc_cf = ceph_alloc_cap_flush(); |
1399 | if (!prealloc_cf) | |
1400 | return -ENOMEM; | |
1401 | ||
a5cd74ad | 1402 | retry_snap: |
5955102c | 1403 | inode_lock(inode); |
03d254ed | 1404 | |
03d254ed | 1405 | /* We can write back this queue in page reclaim */ |
de1414a6 | 1406 | current->backing_dev_info = inode_to_bdi(inode); |
03d254ed | 1407 | |
55b0b31c YZ |
1408 | if (iocb->ki_flags & IOCB_APPEND) { |
1409 | err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); | |
1410 | if (err < 0) | |
1411 | goto out; | |
1412 | } | |
1413 | ||
3309dd04 AV |
1414 | err = generic_write_checks(iocb, from); |
1415 | if (err <= 0) | |
03d254ed YZ |
1416 | goto out; |
1417 | ||
3309dd04 AV |
1418 | pos = iocb->ki_pos; |
1419 | count = iov_iter_count(from); | |
2b83845f LH |
1420 | if (ceph_quota_is_max_bytes_exceeded(inode, pos + count)) { |
1421 | err = -EDQUOT; | |
1422 | goto out; | |
1423 | } | |
1424 | ||
5fa8e0a1 | 1425 | err = file_remove_privs(file); |
03d254ed YZ |
1426 | if (err) |
1427 | goto out; | |
1428 | ||
1429 | err = file_update_time(file); | |
1430 | if (err) | |
1431 | goto out; | |
1432 | ||
28127bdd YZ |
1433 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
1434 | err = ceph_uninline_data(file, NULL); | |
1435 | if (err < 0) | |
1436 | goto out; | |
1437 | } | |
1438 | ||
26544c62 | 1439 | /* FIXME: not complete since it doesn't account for being at quota */ |
b7ec35b3 | 1440 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL)) { |
03d254ed | 1441 | err = -ENOSPC; |
6070e0c1 YZ |
1442 | goto out; |
1443 | } | |
03d254ed | 1444 | |
ac7f29bf | 1445 | dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n", |
99c88e69 | 1446 | inode, ceph_vinop(inode), pos, count, i_size_read(inode)); |
7971bd92 SW |
1447 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
1448 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1449 | else | |
1450 | want = CEPH_CAP_FILE_BUFFER; | |
03d254ed | 1451 | got = 0; |
3738daa6 YZ |
1452 | err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count, |
1453 | &got, NULL); | |
03d254ed | 1454 | if (err < 0) |
37505d57 | 1455 | goto out; |
124e68e7 | 1456 | |
ac7f29bf | 1457 | dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n", |
03d254ed | 1458 | inode, ceph_vinop(inode), pos, count, ceph_cap_string(got)); |
7971bd92 SW |
1459 | |
1460 | if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || | |
26544c62 JL |
1461 | (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC) || |
1462 | (ci->i_ceph_flags & CEPH_I_ERROR_WRITE)) { | |
5dda377c | 1463 | struct ceph_snap_context *snapc; |
4908b822 | 1464 | struct iov_iter data; |
5955102c | 1465 | inode_unlock(inode); |
5dda377c YZ |
1466 | |
1467 | spin_lock(&ci->i_ceph_lock); | |
1468 | if (__ceph_have_pending_cap_snap(ci)) { | |
1469 | struct ceph_cap_snap *capsnap = | |
1470 | list_last_entry(&ci->i_cap_snaps, | |
1471 | struct ceph_cap_snap, | |
1472 | ci_item); | |
1473 | snapc = ceph_get_snap_context(capsnap->context); | |
1474 | } else { | |
1475 | BUG_ON(!ci->i_head_snapc); | |
1476 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
1477 | } | |
1478 | spin_unlock(&ci->i_ceph_lock); | |
1479 | ||
4908b822 AV |
1480 | /* we might need to revert back to that point */ |
1481 | data = *from; | |
2ba48ce5 | 1482 | if (iocb->ki_flags & IOCB_DIRECT) |
c8fe9b17 YZ |
1483 | written = ceph_direct_read_write(iocb, &data, snapc, |
1484 | &prealloc_cf); | |
e8344e66 | 1485 | else |
5dda377c | 1486 | written = ceph_sync_write(iocb, &data, pos, snapc); |
4908b822 AV |
1487 | if (written > 0) |
1488 | iov_iter_advance(from, written); | |
5dda377c | 1489 | ceph_put_snap_context(snapc); |
7971bd92 | 1490 | } else { |
b0d7c223 YZ |
1491 | /* |
1492 | * No need to acquire the i_truncate_mutex. Because | |
1493 | * the MDS revokes Fwb caps before sending truncate | |
1494 | * message to us. We can't get Fwb cap while there | |
1495 | * are pending vmtruncate. So write and vmtruncate | |
1496 | * can not run at the same time | |
1497 | */ | |
4908b822 | 1498 | written = generic_perform_write(file, from, pos); |
aec605f4 AV |
1499 | if (likely(written >= 0)) |
1500 | iocb->ki_pos = pos + written; | |
5955102c | 1501 | inode_unlock(inode); |
7971bd92 | 1502 | } |
d8de9ab6 | 1503 | |
03d254ed | 1504 | if (written >= 0) { |
fca65b4a | 1505 | int dirty; |
1ab302a0 | 1506 | |
be655596 | 1507 | spin_lock(&ci->i_ceph_lock); |
28127bdd | 1508 | ci->i_inline_version = CEPH_INLINE_NONE; |
f66fd9f0 YZ |
1509 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
1510 | &prealloc_cf); | |
be655596 | 1511 | spin_unlock(&ci->i_ceph_lock); |
fca65b4a SW |
1512 | if (dirty) |
1513 | __mark_inode_dirty(inode, dirty); | |
1ab302a0 LH |
1514 | if (ceph_quota_is_max_bytes_approaching(inode, iocb->ki_pos)) |
1515 | ceph_check_caps(ci, CHECK_CAPS_NODELAY, NULL); | |
124e68e7 | 1516 | } |
7971bd92 | 1517 | |
124e68e7 | 1518 | dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n", |
4908b822 | 1519 | inode, ceph_vinop(inode), pos, (unsigned)count, |
7971bd92 | 1520 | ceph_cap_string(got)); |
124e68e7 | 1521 | ceph_put_cap_refs(ci, got); |
7971bd92 | 1522 | |
a5cd74ad YZ |
1523 | if (written == -EOLDSNAPC) { |
1524 | dout("aio_write %p %llx.%llx %llu~%u" "got EOLDSNAPC, retrying\n", | |
1525 | inode, ceph_vinop(inode), pos, (unsigned)count); | |
1526 | goto retry_snap; | |
1527 | } | |
1528 | ||
6aa657c8 | 1529 | if (written >= 0) { |
b7ec35b3 | 1530 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_NEARFULL)) |
6aa657c8 | 1531 | iocb->ki_flags |= IOCB_DSYNC; |
6aa657c8 | 1532 | written = generic_write_sync(iocb, written); |
6070e0c1 | 1533 | } |
03d254ed | 1534 | |
2f75e9e1 SW |
1535 | goto out_unlocked; |
1536 | ||
03d254ed | 1537 | out: |
5955102c | 1538 | inode_unlock(inode); |
2f75e9e1 | 1539 | out_unlocked: |
f66fd9f0 | 1540 | ceph_free_cap_flush(prealloc_cf); |
03d254ed | 1541 | current->backing_dev_info = NULL; |
03d254ed | 1542 | return written ? written : err; |
124e68e7 SW |
1543 | } |
1544 | ||
1545 | /* | |
1546 | * llseek. be sure to verify file size on SEEK_END. | |
1547 | */ | |
965c8e59 | 1548 | static loff_t ceph_llseek(struct file *file, loff_t offset, int whence) |
124e68e7 SW |
1549 | { |
1550 | struct inode *inode = file->f_mapping->host; | |
99c88e69 | 1551 | loff_t i_size; |
955818cd | 1552 | loff_t ret; |
124e68e7 | 1553 | |
5955102c | 1554 | inode_lock(inode); |
6a82c47a | 1555 | |
965c8e59 | 1556 | if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { |
508b32d8 | 1557 | ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); |
955818cd | 1558 | if (ret < 0) |
124e68e7 | 1559 | goto out; |
06222e49 JB |
1560 | } |
1561 | ||
99c88e69 | 1562 | i_size = i_size_read(inode); |
965c8e59 | 1563 | switch (whence) { |
06222e49 | 1564 | case SEEK_END: |
99c88e69 | 1565 | offset += i_size; |
124e68e7 SW |
1566 | break; |
1567 | case SEEK_CUR: | |
1568 | /* | |
1569 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
1570 | * position-querying operation. Avoid rewriting the "same" | |
1571 | * f_pos value back to the file because a concurrent read(), | |
1572 | * write() or lseek() might have altered it | |
1573 | */ | |
1574 | if (offset == 0) { | |
955818cd | 1575 | ret = file->f_pos; |
124e68e7 SW |
1576 | goto out; |
1577 | } | |
1578 | offset += file->f_pos; | |
1579 | break; | |
06222e49 | 1580 | case SEEK_DATA: |
397f2389 | 1581 | if (offset < 0 || offset >= i_size) { |
06222e49 JB |
1582 | ret = -ENXIO; |
1583 | goto out; | |
1584 | } | |
1585 | break; | |
1586 | case SEEK_HOLE: | |
397f2389 | 1587 | if (offset < 0 || offset >= i_size) { |
06222e49 JB |
1588 | ret = -ENXIO; |
1589 | goto out; | |
1590 | } | |
99c88e69 | 1591 | offset = i_size; |
06222e49 | 1592 | break; |
124e68e7 SW |
1593 | } |
1594 | ||
955818cd | 1595 | ret = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
124e68e7 SW |
1596 | |
1597 | out: | |
5955102c | 1598 | inode_unlock(inode); |
955818cd | 1599 | return ret; |
124e68e7 SW |
1600 | } |
1601 | ||
ad7a60de LW |
1602 | static inline void ceph_zero_partial_page( |
1603 | struct inode *inode, loff_t offset, unsigned size) | |
1604 | { | |
1605 | struct page *page; | |
09cbfeaf | 1606 | pgoff_t index = offset >> PAGE_SHIFT; |
ad7a60de LW |
1607 | |
1608 | page = find_lock_page(inode->i_mapping, index); | |
1609 | if (page) { | |
1610 | wait_on_page_writeback(page); | |
09cbfeaf | 1611 | zero_user(page, offset & (PAGE_SIZE - 1), size); |
ad7a60de | 1612 | unlock_page(page); |
09cbfeaf | 1613 | put_page(page); |
ad7a60de LW |
1614 | } |
1615 | } | |
1616 | ||
1617 | static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset, | |
1618 | loff_t length) | |
1619 | { | |
09cbfeaf | 1620 | loff_t nearly = round_up(offset, PAGE_SIZE); |
ad7a60de LW |
1621 | if (offset < nearly) { |
1622 | loff_t size = nearly - offset; | |
1623 | if (length < size) | |
1624 | size = length; | |
1625 | ceph_zero_partial_page(inode, offset, size); | |
1626 | offset += size; | |
1627 | length -= size; | |
1628 | } | |
09cbfeaf KS |
1629 | if (length >= PAGE_SIZE) { |
1630 | loff_t size = round_down(length, PAGE_SIZE); | |
ad7a60de LW |
1631 | truncate_pagecache_range(inode, offset, offset + size - 1); |
1632 | offset += size; | |
1633 | length -= size; | |
1634 | } | |
1635 | if (length) | |
1636 | ceph_zero_partial_page(inode, offset, length); | |
1637 | } | |
1638 | ||
1639 | static int ceph_zero_partial_object(struct inode *inode, | |
1640 | loff_t offset, loff_t *length) | |
1641 | { | |
1642 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1643 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
1644 | struct ceph_osd_request *req; | |
1645 | int ret = 0; | |
1646 | loff_t zero = 0; | |
1647 | int op; | |
1648 | ||
1649 | if (!length) { | |
1650 | op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE; | |
1651 | length = &zero; | |
1652 | } else { | |
1653 | op = CEPH_OSD_OP_ZERO; | |
1654 | } | |
1655 | ||
1656 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1657 | ceph_vino(inode), | |
1658 | offset, length, | |
715e4cd4 | 1659 | 0, 1, op, |
54ea0046 | 1660 | CEPH_OSD_FLAG_WRITE, |
ad7a60de LW |
1661 | NULL, 0, 0, false); |
1662 | if (IS_ERR(req)) { | |
1663 | ret = PTR_ERR(req); | |
1664 | goto out; | |
1665 | } | |
1666 | ||
bb873b53 | 1667 | req->r_mtime = inode->i_mtime; |
ad7a60de LW |
1668 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
1669 | if (!ret) { | |
1670 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1671 | if (ret == -ENOENT) | |
1672 | ret = 0; | |
1673 | } | |
1674 | ceph_osdc_put_request(req); | |
1675 | ||
1676 | out: | |
1677 | return ret; | |
1678 | } | |
1679 | ||
1680 | static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length) | |
1681 | { | |
1682 | int ret = 0; | |
1683 | struct ceph_inode_info *ci = ceph_inode(inode); | |
7627151e YZ |
1684 | s32 stripe_unit = ci->i_layout.stripe_unit; |
1685 | s32 stripe_count = ci->i_layout.stripe_count; | |
1686 | s32 object_size = ci->i_layout.object_size; | |
b314a90d SW |
1687 | u64 object_set_size = object_size * stripe_count; |
1688 | u64 nearly, t; | |
1689 | ||
1690 | /* round offset up to next period boundary */ | |
1691 | nearly = offset + object_set_size - 1; | |
1692 | t = nearly; | |
1693 | nearly -= do_div(t, object_set_size); | |
ad7a60de | 1694 | |
ad7a60de LW |
1695 | while (length && offset < nearly) { |
1696 | loff_t size = length; | |
1697 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1698 | if (ret < 0) | |
1699 | return ret; | |
1700 | offset += size; | |
1701 | length -= size; | |
1702 | } | |
1703 | while (length >= object_set_size) { | |
1704 | int i; | |
1705 | loff_t pos = offset; | |
1706 | for (i = 0; i < stripe_count; ++i) { | |
1707 | ret = ceph_zero_partial_object(inode, pos, NULL); | |
1708 | if (ret < 0) | |
1709 | return ret; | |
1710 | pos += stripe_unit; | |
1711 | } | |
1712 | offset += object_set_size; | |
1713 | length -= object_set_size; | |
1714 | } | |
1715 | while (length) { | |
1716 | loff_t size = length; | |
1717 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1718 | if (ret < 0) | |
1719 | return ret; | |
1720 | offset += size; | |
1721 | length -= size; | |
1722 | } | |
1723 | return ret; | |
1724 | } | |
1725 | ||
1726 | static long ceph_fallocate(struct file *file, int mode, | |
1727 | loff_t offset, loff_t length) | |
1728 | { | |
1729 | struct ceph_file_info *fi = file->private_data; | |
aa8b60e0 | 1730 | struct inode *inode = file_inode(file); |
ad7a60de LW |
1731 | struct ceph_inode_info *ci = ceph_inode(inode); |
1732 | struct ceph_osd_client *osdc = | |
1733 | &ceph_inode_to_client(inode)->client->osdc; | |
f66fd9f0 | 1734 | struct ceph_cap_flush *prealloc_cf; |
ad7a60de LW |
1735 | int want, got = 0; |
1736 | int dirty; | |
1737 | int ret = 0; | |
1738 | loff_t endoff = 0; | |
1739 | loff_t size; | |
1740 | ||
494d77bf YZ |
1741 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
1742 | return -EOPNOTSUPP; | |
1743 | ||
ad7a60de LW |
1744 | if (!S_ISREG(inode->i_mode)) |
1745 | return -EOPNOTSUPP; | |
1746 | ||
f66fd9f0 YZ |
1747 | prealloc_cf = ceph_alloc_cap_flush(); |
1748 | if (!prealloc_cf) | |
1749 | return -ENOMEM; | |
1750 | ||
5955102c | 1751 | inode_lock(inode); |
ad7a60de LW |
1752 | |
1753 | if (ceph_snap(inode) != CEPH_NOSNAP) { | |
1754 | ret = -EROFS; | |
1755 | goto unlock; | |
1756 | } | |
1757 | ||
2b83845f LH |
1758 | if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) && |
1759 | ceph_quota_is_max_bytes_exceeded(inode, offset + length)) { | |
1760 | ret = -EDQUOT; | |
1761 | goto unlock; | |
1762 | } | |
1763 | ||
b7ec35b3 ID |
1764 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) && |
1765 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
ad7a60de LW |
1766 | ret = -ENOSPC; |
1767 | goto unlock; | |
1768 | } | |
1769 | ||
28127bdd YZ |
1770 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
1771 | ret = ceph_uninline_data(file, NULL); | |
1772 | if (ret < 0) | |
1773 | goto unlock; | |
1774 | } | |
1775 | ||
ad7a60de | 1776 | size = i_size_read(inode); |
42c99fc4 | 1777 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
ad7a60de | 1778 | endoff = offset + length; |
42c99fc4 LH |
1779 | ret = inode_newsize_ok(inode, endoff); |
1780 | if (ret) | |
1781 | goto unlock; | |
1782 | } | |
ad7a60de LW |
1783 | |
1784 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1785 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1786 | else | |
1787 | want = CEPH_CAP_FILE_BUFFER; | |
1788 | ||
3738daa6 | 1789 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL); |
ad7a60de LW |
1790 | if (ret < 0) |
1791 | goto unlock; | |
1792 | ||
1793 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1794 | if (offset < size) | |
1795 | ceph_zero_pagecache_range(inode, offset, length); | |
1796 | ret = ceph_zero_objects(inode, offset, length); | |
1797 | } else if (endoff > size) { | |
1798 | truncate_pagecache_range(inode, size, -1); | |
1799 | if (ceph_inode_set_size(inode, endoff)) | |
1800 | ceph_check_caps(ceph_inode(inode), | |
1801 | CHECK_CAPS_AUTHONLY, NULL); | |
1802 | } | |
1803 | ||
1804 | if (!ret) { | |
1805 | spin_lock(&ci->i_ceph_lock); | |
28127bdd | 1806 | ci->i_inline_version = CEPH_INLINE_NONE; |
f66fd9f0 YZ |
1807 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
1808 | &prealloc_cf); | |
ad7a60de LW |
1809 | spin_unlock(&ci->i_ceph_lock); |
1810 | if (dirty) | |
1811 | __mark_inode_dirty(inode, dirty); | |
1ab302a0 LH |
1812 | if ((endoff > size) && |
1813 | ceph_quota_is_max_bytes_approaching(inode, endoff)) | |
1814 | ceph_check_caps(ci, CHECK_CAPS_NODELAY, NULL); | |
ad7a60de LW |
1815 | } |
1816 | ||
1817 | ceph_put_cap_refs(ci, got); | |
1818 | unlock: | |
5955102c | 1819 | inode_unlock(inode); |
f66fd9f0 | 1820 | ceph_free_cap_flush(prealloc_cf); |
ad7a60de LW |
1821 | return ret; |
1822 | } | |
1823 | ||
124e68e7 SW |
1824 | const struct file_operations ceph_file_fops = { |
1825 | .open = ceph_open, | |
1826 | .release = ceph_release, | |
1827 | .llseek = ceph_llseek, | |
3644424d | 1828 | .read_iter = ceph_read_iter, |
4908b822 | 1829 | .write_iter = ceph_write_iter, |
124e68e7 SW |
1830 | .mmap = ceph_mmap, |
1831 | .fsync = ceph_fsync, | |
40819f6f GF |
1832 | .lock = ceph_lock, |
1833 | .flock = ceph_flock, | |
7ce469a5 | 1834 | .splice_read = generic_file_splice_read, |
3551dd79 | 1835 | .splice_write = iter_file_splice_write, |
124e68e7 SW |
1836 | .unlocked_ioctl = ceph_ioctl, |
1837 | .compat_ioctl = ceph_ioctl, | |
ad7a60de | 1838 | .fallocate = ceph_fallocate, |
124e68e7 SW |
1839 | }; |
1840 |