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 | ||
194 | /* | |
124e68e7 SW |
195 | * If we already have the requisite capabilities, we can satisfy |
196 | * the open request locally (no need to request new caps from the | |
197 | * MDS). We do, however, need to inform the MDS (asynchronously) | |
198 | * if our wanted caps set expands. | |
199 | */ | |
200 | int ceph_open(struct inode *inode, struct file *file) | |
201 | { | |
202 | struct ceph_inode_info *ci = ceph_inode(inode); | |
3d14c5d2 YS |
203 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
204 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
205 | struct ceph_mds_request *req; |
206 | struct ceph_file_info *cf = file->private_data; | |
124e68e7 SW |
207 | int err; |
208 | int flags, fmode, wanted; | |
209 | ||
210 | if (cf) { | |
211 | dout("open file %p is already opened\n", file); | |
212 | return 0; | |
213 | } | |
214 | ||
215 | /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */ | |
216 | flags = file->f_flags & ~(O_CREAT|O_EXCL); | |
217 | if (S_ISDIR(inode->i_mode)) | |
218 | flags = O_DIRECTORY; /* mds likes to know */ | |
219 | ||
220 | dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode, | |
221 | ceph_vinop(inode), file, flags, file->f_flags); | |
222 | fmode = ceph_flags_to_mode(flags); | |
223 | wanted = ceph_caps_for_mode(fmode); | |
224 | ||
225 | /* snapped files are read-only */ | |
226 | if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE)) | |
227 | return -EROFS; | |
228 | ||
229 | /* trivially open snapdir */ | |
230 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
be655596 | 231 | spin_lock(&ci->i_ceph_lock); |
124e68e7 | 232 | __ceph_get_fmode(ci, fmode); |
be655596 | 233 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
234 | return ceph_init_file(inode, file, fmode); |
235 | } | |
236 | ||
237 | /* | |
7421ab80 SW |
238 | * No need to block if we have caps on the auth MDS (for |
239 | * write) or any MDS (for read). Update wanted set | |
124e68e7 SW |
240 | * asynchronously. |
241 | */ | |
be655596 | 242 | spin_lock(&ci->i_ceph_lock); |
7421ab80 SW |
243 | if (__ceph_is_any_real_caps(ci) && |
244 | (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) { | |
124e68e7 SW |
245 | int mds_wanted = __ceph_caps_mds_wanted(ci); |
246 | int issued = __ceph_caps_issued(ci, NULL); | |
247 | ||
248 | dout("open %p fmode %d want %s issued %s using existing\n", | |
249 | inode, fmode, ceph_cap_string(wanted), | |
250 | ceph_cap_string(issued)); | |
251 | __ceph_get_fmode(ci, fmode); | |
be655596 | 252 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
253 | |
254 | /* adjust wanted? */ | |
255 | if ((issued & wanted) != wanted && | |
256 | (mds_wanted & wanted) != wanted && | |
257 | ceph_snap(inode) != CEPH_SNAPDIR) | |
258 | ceph_check_caps(ci, 0, NULL); | |
259 | ||
260 | return ceph_init_file(inode, file, fmode); | |
261 | } else if (ceph_snap(inode) != CEPH_NOSNAP && | |
262 | (ci->i_snap_caps & wanted) == wanted) { | |
263 | __ceph_get_fmode(ci, fmode); | |
be655596 | 264 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
265 | return ceph_init_file(inode, file, fmode); |
266 | } | |
99ccbd22 | 267 | |
be655596 | 268 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
269 | |
270 | dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted)); | |
271 | req = prepare_open_request(inode->i_sb, flags, 0); | |
272 | if (IS_ERR(req)) { | |
273 | err = PTR_ERR(req); | |
274 | goto out; | |
275 | } | |
70b666c3 SW |
276 | req->r_inode = inode; |
277 | ihold(inode); | |
99ccbd22 | 278 | |
124e68e7 | 279 | req->r_num_caps = 1; |
e36d571d | 280 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
124e68e7 SW |
281 | if (!err) |
282 | err = ceph_init_file(inode, file, req->r_fmode); | |
283 | ceph_mdsc_put_request(req); | |
284 | dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode)); | |
285 | out: | |
286 | return err; | |
287 | } | |
288 | ||
289 | ||
290 | /* | |
5ef50c3b SW |
291 | * Do a lookup + open with a single request. If we get a non-existent |
292 | * file or symlink, return 1 so the VFS can retry. | |
124e68e7 | 293 | */ |
5ef50c3b | 294 | int ceph_atomic_open(struct inode *dir, struct dentry *dentry, |
30d90494 | 295 | struct file *file, unsigned flags, umode_t mode, |
d9585277 | 296 | int *opened) |
124e68e7 | 297 | { |
3d14c5d2 YS |
298 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
299 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 | 300 | struct ceph_mds_request *req; |
5ef50c3b | 301 | struct dentry *dn; |
b1ee94aa | 302 | struct ceph_acls_info acls = {}; |
315f2408 | 303 | int mask; |
124e68e7 | 304 | int err; |
124e68e7 | 305 | |
a455589f AV |
306 | dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n", |
307 | dir, dentry, dentry, | |
5ef50c3b SW |
308 | d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode); |
309 | ||
310 | if (dentry->d_name.len > NAME_MAX) | |
311 | return -ENAMETOOLONG; | |
312 | ||
313 | err = ceph_init_dentry(dentry); | |
314 | if (err < 0) | |
315 | return err; | |
124e68e7 | 316 | |
b1ee94aa YZ |
317 | if (flags & O_CREAT) { |
318 | err = ceph_pre_init_acls(dir, &mode, &acls); | |
319 | if (err < 0) | |
320 | return err; | |
321 | } | |
322 | ||
124e68e7 SW |
323 | /* do the open */ |
324 | req = prepare_open_request(dir->i_sb, flags, mode); | |
b1ee94aa YZ |
325 | if (IS_ERR(req)) { |
326 | err = PTR_ERR(req); | |
327 | goto out_acl; | |
328 | } | |
124e68e7 SW |
329 | req->r_dentry = dget(dentry); |
330 | req->r_num_caps = 2; | |
331 | if (flags & O_CREAT) { | |
332 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
333 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
b1ee94aa YZ |
334 | if (acls.pagelist) { |
335 | req->r_pagelist = acls.pagelist; | |
336 | acls.pagelist = NULL; | |
337 | } | |
124e68e7 | 338 | } |
315f2408 YZ |
339 | |
340 | mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED; | |
341 | if (ceph_security_xattr_wanted(dir)) | |
342 | mask |= CEPH_CAP_XATTR_SHARED; | |
343 | req->r_args.open.mask = cpu_to_le32(mask); | |
344 | ||
124e68e7 | 345 | req->r_locked_dir = dir; /* caller holds dir->i_mutex */ |
acda7657 SW |
346 | err = ceph_mdsc_do_request(mdsc, |
347 | (flags & (O_CREAT|O_TRUNC)) ? dir : NULL, | |
348 | req); | |
bf91c315 | 349 | err = ceph_handle_snapdir(req, dentry, err); |
79aec984 | 350 | if (err) |
b1ee94aa | 351 | goto out_req; |
79aec984 | 352 | |
a43137f7 | 353 | if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry) |
124e68e7 | 354 | err = ceph_handle_notrace_create(dir, dentry); |
2d83bde9 | 355 | |
5ef50c3b SW |
356 | if (d_unhashed(dentry)) { |
357 | dn = ceph_finish_lookup(req, dentry, err); | |
358 | if (IS_ERR(dn)) | |
359 | err = PTR_ERR(dn); | |
360 | } else { | |
361 | /* we were given a hashed negative dentry */ | |
362 | dn = NULL; | |
363 | } | |
364 | if (err) | |
b1ee94aa | 365 | goto out_req; |
2b0143b5 | 366 | if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) { |
5ef50c3b SW |
367 | /* make vfs retry on splice, ENOENT, or symlink */ |
368 | dout("atomic_open finish_no_open on dn %p\n", dn); | |
369 | err = finish_no_open(file, dn); | |
370 | } else { | |
371 | dout("atomic_open finish_open on dn %p\n", dn); | |
6e8575fa | 372 | if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) { |
2b0143b5 | 373 | ceph_init_inode_acls(d_inode(dentry), &acls); |
6e8575fa SL |
374 | *opened |= FILE_CREATED; |
375 | } | |
5ef50c3b SW |
376 | err = finish_open(file, dentry, ceph_open, opened); |
377 | } | |
b1ee94aa | 378 | out_req: |
ab866549 YZ |
379 | if (!req->r_err && req->r_target_inode) |
380 | ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode); | |
5ef50c3b | 381 | ceph_mdsc_put_request(req); |
b1ee94aa YZ |
382 | out_acl: |
383 | ceph_release_acls_info(&acls); | |
5ef50c3b | 384 | dout("atomic_open result=%d\n", err); |
d9585277 | 385 | return err; |
124e68e7 SW |
386 | } |
387 | ||
388 | int ceph_release(struct inode *inode, struct file *file) | |
389 | { | |
390 | struct ceph_inode_info *ci = ceph_inode(inode); | |
391 | struct ceph_file_info *cf = file->private_data; | |
392 | ||
393 | dout("release inode %p file %p\n", inode, file); | |
394 | ceph_put_fmode(ci, cf->fmode); | |
395 | if (cf->last_readdir) | |
396 | ceph_mdsc_put_request(cf->last_readdir); | |
397 | kfree(cf->last_name); | |
398 | kfree(cf->dir_info); | |
124e68e7 | 399 | kmem_cache_free(ceph_file_cachep, cf); |
195d3ce2 SW |
400 | |
401 | /* wake up anyone waiting for caps on this inode */ | |
03066f23 | 402 | wake_up_all(&ci->i_cap_wq); |
124e68e7 SW |
403 | return 0; |
404 | } | |
405 | ||
83701246 | 406 | enum { |
c8fe9b17 YZ |
407 | HAVE_RETRIED = 1, |
408 | CHECK_EOF = 2, | |
409 | READ_INLINE = 3, | |
83701246 YZ |
410 | }; |
411 | ||
124e68e7 SW |
412 | /* |
413 | * Read a range of bytes striped over one or more objects. Iterate over | |
414 | * objects we stripe over. (That's not atomic, but good enough for now.) | |
415 | * | |
416 | * If we get a short result from the OSD, check against i_size; we need to | |
417 | * only return a short read to the caller if we hit EOF. | |
418 | */ | |
419 | static int striped_read(struct inode *inode, | |
420 | u64 off, u64 len, | |
6a026589 | 421 | struct page **pages, int num_pages, |
c8fe9b17 | 422 | int *checkeof) |
124e68e7 | 423 | { |
3d14c5d2 | 424 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
124e68e7 | 425 | struct ceph_inode_info *ci = ceph_inode(inode); |
688bac46 | 426 | u64 pos, this_len, left; |
99c88e69 | 427 | loff_t i_size; |
c8fe9b17 YZ |
428 | int page_align, pages_left; |
429 | int read, ret; | |
124e68e7 | 430 | struct page **page_pos; |
124e68e7 SW |
431 | bool hit_stripe, was_short; |
432 | ||
433 | /* | |
434 | * we may need to do multiple reads. not atomic, unfortunately. | |
435 | */ | |
436 | pos = off; | |
437 | left = len; | |
438 | page_pos = pages; | |
439 | pages_left = num_pages; | |
440 | read = 0; | |
441 | ||
442 | more: | |
c8fe9b17 | 443 | page_align = pos & ~PAGE_MASK; |
124e68e7 | 444 | this_len = left; |
3d14c5d2 | 445 | ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), |
124e68e7 SW |
446 | &ci->i_layout, pos, &this_len, |
447 | ci->i_truncate_seq, | |
448 | ci->i_truncate_size, | |
b7495fc2 | 449 | page_pos, pages_left, page_align); |
124e68e7 SW |
450 | if (ret == -ENOENT) |
451 | ret = 0; | |
0e98728f SW |
452 | hit_stripe = this_len < left; |
453 | was_short = ret >= 0 && ret < this_len; | |
688bac46 | 454 | dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read, |
124e68e7 SW |
455 | ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : ""); |
456 | ||
99c88e69 | 457 | i_size = i_size_read(inode); |
02ae66d8 | 458 | if (ret >= 0) { |
459 | int didpages; | |
99c88e69 YZ |
460 | if (was_short && (pos + ret < i_size)) { |
461 | int zlen = min(this_len - ret, i_size - pos - ret); | |
c8fe9b17 | 462 | int zoff = (off & ~PAGE_MASK) + read + ret; |
02ae66d8 | 463 | dout(" zero gap %llu to %llu\n", |
1487a688 YZ |
464 | pos + ret, pos + ret + zlen); |
465 | ceph_zero_page_vector_range(zoff, zlen, pages); | |
466 | ret += zlen; | |
124e68e7 | 467 | } |
02ae66d8 | 468 | |
09cbfeaf | 469 | didpages = (page_align + ret) >> PAGE_SHIFT; |
124e68e7 SW |
470 | pos += ret; |
471 | read = pos - off; | |
472 | left -= ret; | |
473 | page_pos += didpages; | |
474 | pages_left -= didpages; | |
475 | ||
02ae66d8 | 476 | /* hit stripe and need continue*/ |
99c88e69 | 477 | if (left && hit_stripe && pos < i_size) |
124e68e7 SW |
478 | goto more; |
479 | } | |
480 | ||
ee7289bf | 481 | if (read > 0) { |
02ae66d8 | 482 | ret = read; |
c3cd6283 | 483 | /* did we bounce off eof? */ |
99c88e69 | 484 | if (pos + left > i_size) |
83701246 | 485 | *checkeof = CHECK_EOF; |
124e68e7 SW |
486 | } |
487 | ||
124e68e7 SW |
488 | dout("striped_read returns %d\n", ret); |
489 | return ret; | |
490 | } | |
491 | ||
492 | /* | |
493 | * Completely synchronous read and write methods. Direct from __user | |
494 | * buffer to osd, or directly to user pages (if O_DIRECT). | |
495 | * | |
496 | * If the read spans object boundary, just do multiple reads. | |
497 | */ | |
8eb4efb0 | 498 | static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i, |
499 | int *checkeof) | |
124e68e7 | 500 | { |
8eb4efb0 | 501 | struct file *file = iocb->ki_filp; |
496ad9aa | 502 | struct inode *inode = file_inode(file); |
124e68e7 | 503 | struct page **pages; |
8eb4efb0 | 504 | u64 off = iocb->ki_pos; |
ab226e21 | 505 | int num_pages, ret; |
2b777c9d | 506 | size_t len = iov_iter_count(i); |
124e68e7 | 507 | |
8eb4efb0 | 508 | dout("sync_read on file %p %llu~%u %s\n", file, off, |
509 | (unsigned)len, | |
124e68e7 | 510 | (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); |
d0d0db22 YZ |
511 | |
512 | if (!len) | |
513 | return 0; | |
e98b6fed SW |
514 | /* |
515 | * flush any page cache pages in this range. this | |
516 | * will make concurrent normal and sync io slow, | |
517 | * but it will at least behave sensibly when they are | |
518 | * in sequence. | |
519 | */ | |
8eb4efb0 | 520 | ret = filemap_write_and_wait_range(inode->i_mapping, off, |
521 | off + len); | |
29065a51 | 522 | if (ret < 0) |
8eb4efb0 | 523 | return ret; |
29065a51 | 524 | |
c8fe9b17 YZ |
525 | num_pages = calc_pages_for(off, len); |
526 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); | |
527 | if (IS_ERR(pages)) | |
528 | return PTR_ERR(pages); | |
529 | ret = striped_read(inode, off, len, pages, | |
530 | num_pages, checkeof); | |
531 | if (ret > 0) { | |
532 | int l, k = 0; | |
533 | size_t left = ret; | |
534 | ||
535 | while (left) { | |
536 | size_t page_off = off & ~PAGE_MASK; | |
537 | size_t copy = min_t(size_t, left, | |
538 | PAGE_SIZE - page_off); | |
539 | l = copy_page_to_iter(pages[k++], page_off, copy, i); | |
540 | off += l; | |
541 | left -= l; | |
542 | if (l < copy) | |
8eb4efb0 | 543 | break; |
544 | } | |
8eb4efb0 | 545 | } |
c8fe9b17 | 546 | ceph_release_page_vector(pages, num_pages); |
124e68e7 | 547 | |
8eb4efb0 | 548 | if (off > iocb->ki_pos) { |
549 | ret = off - iocb->ki_pos; | |
550 | iocb->ki_pos = off; | |
551 | } | |
124e68e7 | 552 | |
124e68e7 SW |
553 | dout("sync_read result %d\n", ret); |
554 | return ret; | |
555 | } | |
556 | ||
c8fe9b17 YZ |
557 | struct ceph_aio_request { |
558 | struct kiocb *iocb; | |
559 | size_t total_len; | |
560 | int write; | |
561 | int error; | |
562 | struct list_head osd_reqs; | |
563 | unsigned num_reqs; | |
564 | atomic_t pending_reqs; | |
5be0389d | 565 | struct timespec mtime; |
c8fe9b17 YZ |
566 | struct ceph_cap_flush *prealloc_cf; |
567 | }; | |
568 | ||
5be0389d YZ |
569 | struct ceph_aio_work { |
570 | struct work_struct work; | |
571 | struct ceph_osd_request *req; | |
572 | }; | |
573 | ||
574 | static void ceph_aio_retry_work(struct work_struct *work); | |
575 | ||
c8fe9b17 YZ |
576 | static void ceph_aio_complete(struct inode *inode, |
577 | struct ceph_aio_request *aio_req) | |
578 | { | |
579 | struct ceph_inode_info *ci = ceph_inode(inode); | |
580 | int ret; | |
581 | ||
582 | if (!atomic_dec_and_test(&aio_req->pending_reqs)) | |
583 | return; | |
584 | ||
585 | ret = aio_req->error; | |
586 | if (!ret) | |
587 | ret = aio_req->total_len; | |
588 | ||
589 | dout("ceph_aio_complete %p rc %d\n", inode, ret); | |
590 | ||
591 | if (ret >= 0 && aio_req->write) { | |
592 | int dirty; | |
593 | ||
594 | loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len; | |
595 | if (endoff > i_size_read(inode)) { | |
596 | if (ceph_inode_set_size(inode, endoff)) | |
597 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); | |
598 | } | |
599 | ||
600 | spin_lock(&ci->i_ceph_lock); | |
601 | ci->i_inline_version = CEPH_INLINE_NONE; | |
602 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, | |
603 | &aio_req->prealloc_cf); | |
604 | spin_unlock(&ci->i_ceph_lock); | |
605 | if (dirty) | |
606 | __mark_inode_dirty(inode, dirty); | |
607 | ||
608 | } | |
609 | ||
610 | ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR : | |
611 | CEPH_CAP_FILE_RD)); | |
612 | ||
613 | aio_req->iocb->ki_complete(aio_req->iocb, ret, 0); | |
614 | ||
615 | ceph_free_cap_flush(aio_req->prealloc_cf); | |
616 | kfree(aio_req); | |
617 | } | |
618 | ||
619 | static void ceph_aio_complete_req(struct ceph_osd_request *req, | |
620 | struct ceph_msg *msg) | |
621 | { | |
622 | int rc = req->r_result; | |
623 | struct inode *inode = req->r_inode; | |
624 | struct ceph_aio_request *aio_req = req->r_priv; | |
625 | struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0); | |
626 | int num_pages = calc_pages_for((u64)osd_data->alignment, | |
627 | osd_data->length); | |
628 | ||
629 | dout("ceph_aio_complete_req %p rc %d bytes %llu\n", | |
630 | inode, rc, osd_data->length); | |
631 | ||
632 | if (rc == -EOLDSNAPC) { | |
5be0389d YZ |
633 | struct ceph_aio_work *aio_work; |
634 | BUG_ON(!aio_req->write); | |
635 | ||
636 | aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS); | |
637 | if (aio_work) { | |
638 | INIT_WORK(&aio_work->work, ceph_aio_retry_work); | |
639 | aio_work->req = req; | |
640 | queue_work(ceph_inode_to_client(inode)->wb_wq, | |
641 | &aio_work->work); | |
642 | return; | |
643 | } | |
644 | rc = -ENOMEM; | |
645 | } else if (!aio_req->write) { | |
c8fe9b17 YZ |
646 | if (rc == -ENOENT) |
647 | rc = 0; | |
648 | if (rc >= 0 && osd_data->length > rc) { | |
649 | int zoff = osd_data->alignment + rc; | |
650 | int zlen = osd_data->length - rc; | |
651 | /* | |
652 | * If read is satisfied by single OSD request, | |
653 | * it can pass EOF. Otherwise read is within | |
654 | * i_size. | |
655 | */ | |
656 | if (aio_req->num_reqs == 1) { | |
657 | loff_t i_size = i_size_read(inode); | |
658 | loff_t endoff = aio_req->iocb->ki_pos + rc; | |
659 | if (endoff < i_size) | |
660 | zlen = min_t(size_t, zlen, | |
661 | i_size - endoff); | |
662 | aio_req->total_len = rc + zlen; | |
663 | } | |
664 | ||
665 | if (zlen > 0) | |
666 | ceph_zero_page_vector_range(zoff, zlen, | |
667 | osd_data->pages); | |
668 | } | |
669 | } | |
670 | ||
671 | ceph_put_page_vector(osd_data->pages, num_pages, false); | |
672 | ceph_osdc_put_request(req); | |
673 | ||
674 | if (rc < 0) | |
675 | cmpxchg(&aio_req->error, 0, rc); | |
676 | ||
677 | ceph_aio_complete(inode, aio_req); | |
678 | return; | |
679 | } | |
680 | ||
5be0389d YZ |
681 | static void ceph_aio_retry_work(struct work_struct *work) |
682 | { | |
683 | struct ceph_aio_work *aio_work = | |
684 | container_of(work, struct ceph_aio_work, work); | |
685 | struct ceph_osd_request *orig_req = aio_work->req; | |
686 | struct ceph_aio_request *aio_req = orig_req->r_priv; | |
687 | struct inode *inode = orig_req->r_inode; | |
688 | struct ceph_inode_info *ci = ceph_inode(inode); | |
689 | struct ceph_snap_context *snapc; | |
690 | struct ceph_osd_request *req; | |
691 | int ret; | |
692 | ||
693 | spin_lock(&ci->i_ceph_lock); | |
694 | if (__ceph_have_pending_cap_snap(ci)) { | |
695 | struct ceph_cap_snap *capsnap = | |
696 | list_last_entry(&ci->i_cap_snaps, | |
697 | struct ceph_cap_snap, | |
698 | ci_item); | |
699 | snapc = ceph_get_snap_context(capsnap->context); | |
700 | } else { | |
701 | BUG_ON(!ci->i_head_snapc); | |
702 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
703 | } | |
704 | spin_unlock(&ci->i_ceph_lock); | |
705 | ||
706 | req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 2, | |
707 | false, GFP_NOFS); | |
1418bf07 DC |
708 | if (!req) { |
709 | ret = -ENOMEM; | |
5be0389d YZ |
710 | req = orig_req; |
711 | goto out; | |
712 | } | |
713 | ||
714 | req->r_flags = CEPH_OSD_FLAG_ORDERSNAP | | |
715 | CEPH_OSD_FLAG_ONDISK | | |
716 | CEPH_OSD_FLAG_WRITE; | |
717 | req->r_base_oloc = orig_req->r_base_oloc; | |
d30291b9 | 718 | ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid); |
5be0389d | 719 | |
13d1ad16 ID |
720 | ret = ceph_osdc_alloc_messages(req, GFP_NOFS); |
721 | if (ret) { | |
722 | ceph_osdc_put_request(req); | |
723 | req = orig_req; | |
724 | goto out; | |
725 | } | |
726 | ||
5be0389d YZ |
727 | req->r_ops[0] = orig_req->r_ops[0]; |
728 | osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0); | |
729 | ||
730 | ceph_osdc_build_request(req, req->r_ops[0].extent.offset, | |
731 | snapc, CEPH_NOSNAP, &aio_req->mtime); | |
732 | ||
5be0389d YZ |
733 | ceph_osdc_put_request(orig_req); |
734 | ||
735 | req->r_callback = ceph_aio_complete_req; | |
736 | req->r_inode = inode; | |
737 | req->r_priv = aio_req; | |
738 | ||
739 | ret = ceph_osdc_start_request(req->r_osdc, req, false); | |
740 | out: | |
741 | if (ret < 0) { | |
5be0389d YZ |
742 | req->r_result = ret; |
743 | ceph_aio_complete_req(req, NULL); | |
744 | } | |
745 | ||
db6aed70 | 746 | ceph_put_snap_context(snapc); |
5be0389d YZ |
747 | kfree(aio_work); |
748 | } | |
749 | ||
124e68e7 | 750 | /* |
26be8808 AE |
751 | * Write commit request unsafe callback, called to tell us when a |
752 | * request is unsafe (that is, in flight--has been handed to the | |
753 | * messenger to send to its target osd). It is called again when | |
754 | * we've received a response message indicating the request is | |
755 | * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request | |
756 | * is completed early (and unsuccessfully) due to a timeout or | |
757 | * interrupt. | |
758 | * | |
759 | * This is used if we requested both an ACK and ONDISK commit reply | |
760 | * from the OSD. | |
124e68e7 | 761 | */ |
26be8808 | 762 | static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe) |
124e68e7 SW |
763 | { |
764 | struct ceph_inode_info *ci = ceph_inode(req->r_inode); | |
765 | ||
26be8808 AE |
766 | dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid, |
767 | unsafe ? "un" : ""); | |
768 | if (unsafe) { | |
769 | ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR); | |
770 | spin_lock(&ci->i_unsafe_lock); | |
771 | list_add_tail(&req->r_unsafe_item, | |
772 | &ci->i_unsafe_writes); | |
773 | spin_unlock(&ci->i_unsafe_lock); | |
774 | } else { | |
775 | spin_lock(&ci->i_unsafe_lock); | |
776 | list_del_init(&req->r_unsafe_item); | |
777 | spin_unlock(&ci->i_unsafe_lock); | |
778 | ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR); | |
779 | } | |
124e68e7 SW |
780 | } |
781 | ||
e8344e66 | 782 | |
e8344e66 | 783 | static ssize_t |
c8fe9b17 YZ |
784 | ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter, |
785 | struct ceph_snap_context *snapc, | |
786 | struct ceph_cap_flush **pcf) | |
124e68e7 | 787 | { |
e8344e66 | 788 | struct file *file = iocb->ki_filp; |
496ad9aa | 789 | struct inode *inode = file_inode(file); |
124e68e7 | 790 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 | 791 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
acead002 | 792 | struct ceph_vino vino; |
124e68e7 SW |
793 | struct ceph_osd_request *req; |
794 | struct page **pages; | |
c8fe9b17 YZ |
795 | struct ceph_aio_request *aio_req = NULL; |
796 | int num_pages = 0; | |
124e68e7 | 797 | int flags; |
124e68e7 | 798 | int ret; |
8bbd4714 | 799 | struct timespec mtime = current_fs_time(inode->i_sb); |
c8fe9b17 YZ |
800 | size_t count = iov_iter_count(iter); |
801 | loff_t pos = iocb->ki_pos; | |
802 | bool write = iov_iter_rw(iter) == WRITE; | |
124e68e7 | 803 | |
c8fe9b17 | 804 | if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
124e68e7 SW |
805 | return -EROFS; |
806 | ||
c8fe9b17 YZ |
807 | dout("sync_direct_read_write (%s) on file %p %lld~%u\n", |
808 | (write ? "write" : "read"), file, pos, (unsigned)count); | |
124e68e7 | 809 | |
e8344e66 | 810 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); |
29065a51 YS |
811 | if (ret < 0) |
812 | return ret; | |
813 | ||
c8fe9b17 YZ |
814 | if (write) { |
815 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
09cbfeaf KS |
816 | pos >> PAGE_SHIFT, |
817 | (pos + count) >> PAGE_SHIFT); | |
c8fe9b17 YZ |
818 | if (ret < 0) |
819 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
29065a51 | 820 | |
c8fe9b17 YZ |
821 | flags = CEPH_OSD_FLAG_ORDERSNAP | |
822 | CEPH_OSD_FLAG_ONDISK | | |
823 | CEPH_OSD_FLAG_WRITE; | |
824 | } else { | |
825 | flags = CEPH_OSD_FLAG_READ; | |
826 | } | |
124e68e7 | 827 | |
c8fe9b17 YZ |
828 | while (iov_iter_count(iter) > 0) { |
829 | u64 size = dio_get_pagev_size(iter); | |
830 | size_t start = 0; | |
831 | ssize_t len; | |
e8344e66 | 832 | |
e8344e66 | 833 | vino = ceph_vino(inode); |
834 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
c8fe9b17 YZ |
835 | vino, pos, &size, 0, |
836 | /*include a 'startsync' command*/ | |
837 | write ? 2 : 1, | |
838 | write ? CEPH_OSD_OP_WRITE : | |
839 | CEPH_OSD_OP_READ, | |
840 | flags, snapc, | |
e8344e66 | 841 | ci->i_truncate_seq, |
842 | ci->i_truncate_size, | |
843 | false); | |
844 | if (IS_ERR(req)) { | |
845 | ret = PTR_ERR(req); | |
eab87235 | 846 | break; |
e8344e66 | 847 | } |
124e68e7 | 848 | |
c8fe9b17 YZ |
849 | len = size; |
850 | pages = dio_get_pages_alloc(iter, len, &start, &num_pages); | |
b5b98989 | 851 | if (IS_ERR(pages)) { |
64c31311 | 852 | ceph_osdc_put_request(req); |
b5b98989 | 853 | ret = PTR_ERR(pages); |
64c31311 | 854 | break; |
124e68e7 SW |
855 | } |
856 | ||
857 | /* | |
c8fe9b17 YZ |
858 | * To simplify error handling, allow AIO when IO within i_size |
859 | * or IO can be satisfied by single OSD request. | |
124e68e7 | 860 | */ |
c8fe9b17 YZ |
861 | if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) && |
862 | (len == count || pos + count <= i_size_read(inode))) { | |
863 | aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL); | |
864 | if (aio_req) { | |
865 | aio_req->iocb = iocb; | |
866 | aio_req->write = write; | |
867 | INIT_LIST_HEAD(&aio_req->osd_reqs); | |
868 | if (write) { | |
5be0389d | 869 | aio_req->mtime = mtime; |
c8fe9b17 YZ |
870 | swap(aio_req->prealloc_cf, *pcf); |
871 | } | |
872 | } | |
873 | /* ignore error */ | |
874 | } | |
875 | ||
876 | if (write) { | |
877 | /* | |
878 | * throw out any page cache pages in this range. this | |
879 | * may block. | |
880 | */ | |
881 | truncate_inode_pages_range(inode->i_mapping, pos, | |
09cbfeaf | 882 | (pos+len) | (PAGE_SIZE - 1)); |
c8fe9b17 YZ |
883 | |
884 | osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0); | |
885 | } | |
886 | ||
887 | ||
888 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, start, | |
889 | false, false); | |
e8344e66 | 890 | |
e8344e66 | 891 | ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime); |
892 | ||
c8fe9b17 YZ |
893 | if (aio_req) { |
894 | aio_req->total_len += len; | |
895 | aio_req->num_reqs++; | |
896 | atomic_inc(&aio_req->pending_reqs); | |
897 | ||
898 | req->r_callback = ceph_aio_complete_req; | |
899 | req->r_inode = inode; | |
900 | req->r_priv = aio_req; | |
901 | list_add_tail(&req->r_unsafe_item, &aio_req->osd_reqs); | |
902 | ||
903 | pos += len; | |
904 | iov_iter_advance(iter, len); | |
905 | continue; | |
906 | } | |
907 | ||
908 | ret = ceph_osdc_start_request(req->r_osdc, req, false); | |
e8344e66 | 909 | if (!ret) |
910 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
911 | ||
c8fe9b17 YZ |
912 | size = i_size_read(inode); |
913 | if (!write) { | |
914 | if (ret == -ENOENT) | |
915 | ret = 0; | |
916 | if (ret >= 0 && ret < len && pos + ret < size) { | |
917 | int zlen = min_t(size_t, len - ret, | |
918 | size - pos - ret); | |
919 | ceph_zero_page_vector_range(start + ret, zlen, | |
920 | pages); | |
921 | ret += zlen; | |
922 | } | |
923 | if (ret >= 0) | |
924 | len = ret; | |
925 | } | |
926 | ||
e8344e66 | 927 | ceph_put_page_vector(pages, num_pages, false); |
928 | ||
e8344e66 | 929 | ceph_osdc_put_request(req); |
c8fe9b17 | 930 | if (ret < 0) |
e8344e66 | 931 | break; |
64c31311 | 932 | |
c8fe9b17 YZ |
933 | pos += len; |
934 | iov_iter_advance(iter, len); | |
935 | ||
936 | if (!write && pos >= size) | |
e8344e66 | 937 | break; |
64c31311 | 938 | |
c8fe9b17 YZ |
939 | if (write && pos > size) { |
940 | if (ceph_inode_set_size(inode, pos)) | |
64c31311 AV |
941 | ceph_check_caps(ceph_inode(inode), |
942 | CHECK_CAPS_AUTHONLY, | |
943 | NULL); | |
944 | } | |
e8344e66 | 945 | } |
946 | ||
c8fe9b17 YZ |
947 | if (aio_req) { |
948 | if (aio_req->num_reqs == 0) { | |
949 | kfree(aio_req); | |
950 | return ret; | |
951 | } | |
952 | ||
953 | ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR : | |
954 | CEPH_CAP_FILE_RD); | |
955 | ||
956 | while (!list_empty(&aio_req->osd_reqs)) { | |
957 | req = list_first_entry(&aio_req->osd_reqs, | |
958 | struct ceph_osd_request, | |
959 | r_unsafe_item); | |
960 | list_del_init(&req->r_unsafe_item); | |
961 | if (ret >= 0) | |
962 | ret = ceph_osdc_start_request(req->r_osdc, | |
963 | req, false); | |
964 | if (ret < 0) { | |
965 | req->r_result = ret; | |
966 | ceph_aio_complete_req(req, NULL); | |
967 | } | |
968 | } | |
969 | return -EIOCBQUEUED; | |
970 | } | |
971 | ||
972 | if (ret != -EOLDSNAPC && pos > iocb->ki_pos) { | |
973 | ret = pos - iocb->ki_pos; | |
e8344e66 | 974 | iocb->ki_pos = pos; |
e8344e66 | 975 | } |
976 | return ret; | |
977 | } | |
978 | ||
e8344e66 | 979 | /* |
980 | * Synchronous write, straight from __user pointer or user pages. | |
981 | * | |
982 | * If write spans object boundary, just do multiple writes. (For a | |
983 | * correct atomic write, we should e.g. take write locks on all | |
984 | * objects, rollback on failure, etc.) | |
985 | */ | |
06fee30f | 986 | static ssize_t |
5dda377c YZ |
987 | ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos, |
988 | struct ceph_snap_context *snapc) | |
e8344e66 | 989 | { |
990 | struct file *file = iocb->ki_filp; | |
991 | struct inode *inode = file_inode(file); | |
992 | struct ceph_inode_info *ci = ceph_inode(inode); | |
993 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
e8344e66 | 994 | struct ceph_vino vino; |
995 | struct ceph_osd_request *req; | |
996 | struct page **pages; | |
997 | u64 len; | |
998 | int num_pages; | |
999 | int written = 0; | |
1000 | int flags; | |
1001 | int check_caps = 0; | |
1002 | int ret; | |
8bbd4714 | 1003 | struct timespec mtime = current_fs_time(inode->i_sb); |
4908b822 | 1004 | size_t count = iov_iter_count(from); |
e8344e66 | 1005 | |
1006 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) | |
1007 | return -EROFS; | |
1008 | ||
1009 | dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count); | |
1010 | ||
1011 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); | |
1012 | if (ret < 0) | |
1013 | return ret; | |
1014 | ||
1015 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
09cbfeaf KS |
1016 | pos >> PAGE_SHIFT, |
1017 | (pos + count) >> PAGE_SHIFT); | |
e8344e66 | 1018 | if (ret < 0) |
1019 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
1020 | ||
1021 | flags = CEPH_OSD_FLAG_ORDERSNAP | | |
1022 | CEPH_OSD_FLAG_ONDISK | | |
1023 | CEPH_OSD_FLAG_WRITE | | |
1024 | CEPH_OSD_FLAG_ACK; | |
1025 | ||
4908b822 | 1026 | while ((len = iov_iter_count(from)) > 0) { |
e8344e66 | 1027 | size_t left; |
1028 | int n; | |
1029 | ||
e8344e66 | 1030 | vino = ceph_vino(inode); |
1031 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
715e4cd4 | 1032 | vino, pos, &len, 0, 1, |
e8344e66 | 1033 | CEPH_OSD_OP_WRITE, flags, snapc, |
1034 | ci->i_truncate_seq, | |
1035 | ci->i_truncate_size, | |
1036 | false); | |
1037 | if (IS_ERR(req)) { | |
1038 | ret = PTR_ERR(req); | |
eab87235 | 1039 | break; |
e8344e66 | 1040 | } |
1041 | ||
1042 | /* | |
1043 | * write from beginning of first page, | |
1044 | * regardless of io alignment | |
1045 | */ | |
09cbfeaf | 1046 | num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
e8344e66 | 1047 | |
687265e5 | 1048 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
124e68e7 SW |
1049 | if (IS_ERR(pages)) { |
1050 | ret = PTR_ERR(pages); | |
1051 | goto out; | |
1052 | } | |
e8344e66 | 1053 | |
1054 | left = len; | |
1055 | for (n = 0; n < num_pages; n++) { | |
125d725c | 1056 | size_t plen = min_t(size_t, left, PAGE_SIZE); |
4908b822 | 1057 | ret = copy_page_from_iter(pages[n], 0, plen, from); |
e8344e66 | 1058 | if (ret != plen) { |
1059 | ret = -EFAULT; | |
1060 | break; | |
1061 | } | |
1062 | left -= ret; | |
e8344e66 | 1063 | } |
1064 | ||
124e68e7 SW |
1065 | if (ret < 0) { |
1066 | ceph_release_page_vector(pages, num_pages); | |
1067 | goto out; | |
1068 | } | |
1069 | ||
e8344e66 | 1070 | /* get a second commit callback */ |
1071 | req->r_unsafe_callback = ceph_sync_write_unsafe; | |
1072 | req->r_inode = inode; | |
124e68e7 | 1073 | |
e8344e66 | 1074 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, |
1075 | false, true); | |
02ee07d3 | 1076 | |
e8344e66 | 1077 | /* BUG_ON(vino.snap != CEPH_NOSNAP); */ |
1078 | ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime); | |
124e68e7 | 1079 | |
e8344e66 | 1080 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
1081 | if (!ret) | |
1082 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
124e68e7 SW |
1083 | |
1084 | out: | |
e8344e66 | 1085 | ceph_osdc_put_request(req); |
1086 | if (ret == 0) { | |
1087 | pos += len; | |
1088 | written += len; | |
1089 | ||
1090 | if (pos > i_size_read(inode)) { | |
1091 | check_caps = ceph_inode_set_size(inode, pos); | |
1092 | if (check_caps) | |
1093 | ceph_check_caps(ceph_inode(inode), | |
1094 | CHECK_CAPS_AUTHONLY, | |
1095 | NULL); | |
1096 | } | |
1097 | } else | |
1098 | break; | |
1099 | } | |
124e68e7 | 1100 | |
e8344e66 | 1101 | if (ret != -EOLDSNAPC && written > 0) { |
124e68e7 | 1102 | ret = written; |
e8344e66 | 1103 | iocb->ki_pos = pos; |
124e68e7 SW |
1104 | } |
1105 | return ret; | |
1106 | } | |
1107 | ||
1108 | /* | |
1109 | * Wrap generic_file_aio_read with checks for cap bits on the inode. | |
1110 | * Atomically grab references, so that those bits are not released | |
1111 | * back to the MDS mid-read. | |
1112 | * | |
1113 | * Hmm, the sync read case isn't actually async... should it be? | |
1114 | */ | |
3644424d | 1115 | static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) |
124e68e7 SW |
1116 | { |
1117 | struct file *filp = iocb->ki_filp; | |
2962507c | 1118 | struct ceph_file_info *fi = filp->private_data; |
66ee59af | 1119 | size_t len = iov_iter_count(to); |
496ad9aa | 1120 | struct inode *inode = file_inode(filp); |
124e68e7 | 1121 | struct ceph_inode_info *ci = ceph_inode(inode); |
3738daa6 | 1122 | struct page *pinned_page = NULL; |
124e68e7 | 1123 | ssize_t ret; |
2962507c | 1124 | int want, got = 0; |
83701246 | 1125 | int retry_op = 0, read = 0; |
124e68e7 | 1126 | |
6a026589 | 1127 | again: |
8eb4efb0 | 1128 | dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n", |
1129 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode); | |
1130 | ||
2962507c SW |
1131 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
1132 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
1133 | else | |
1134 | want = CEPH_CAP_FILE_CACHE; | |
3738daa6 | 1135 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page); |
124e68e7 | 1136 | if (ret < 0) |
8eb4efb0 | 1137 | return ret; |
124e68e7 | 1138 | |
2962507c | 1139 | if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 || |
2ba48ce5 | 1140 | (iocb->ki_flags & IOCB_DIRECT) || |
8eb4efb0 | 1141 | (fi->flags & CEPH_F_SYNC)) { |
8eb4efb0 | 1142 | |
1143 | dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n", | |
1144 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, | |
1145 | ceph_cap_string(got)); | |
1146 | ||
83701246 | 1147 | if (ci->i_inline_version == CEPH_INLINE_NONE) { |
c8fe9b17 YZ |
1148 | if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) { |
1149 | ret = ceph_direct_read_write(iocb, to, | |
1150 | NULL, NULL); | |
1151 | if (ret >= 0 && ret < len) | |
1152 | retry_op = CHECK_EOF; | |
1153 | } else { | |
1154 | ret = ceph_sync_read(iocb, to, &retry_op); | |
1155 | } | |
83701246 YZ |
1156 | } else { |
1157 | retry_op = READ_INLINE; | |
1158 | } | |
8eb4efb0 | 1159 | } else { |
8eb4efb0 | 1160 | dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n", |
3644424d | 1161 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, |
8eb4efb0 | 1162 | ceph_cap_string(got)); |
124e68e7 | 1163 | |
3644424d | 1164 | ret = generic_file_read_iter(iocb, to); |
8eb4efb0 | 1165 | } |
124e68e7 SW |
1166 | dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", |
1167 | inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); | |
3738daa6 | 1168 | if (pinned_page) { |
09cbfeaf | 1169 | put_page(pinned_page); |
3738daa6 YZ |
1170 | pinned_page = NULL; |
1171 | } | |
124e68e7 | 1172 | ceph_put_cap_refs(ci, got); |
c8fe9b17 | 1173 | if (retry_op > HAVE_RETRIED && ret >= 0) { |
83701246 YZ |
1174 | int statret; |
1175 | struct page *page = NULL; | |
1176 | loff_t i_size; | |
1177 | if (retry_op == READ_INLINE) { | |
687265e5 | 1178 | page = __page_cache_alloc(GFP_KERNEL); |
83701246 YZ |
1179 | if (!page) |
1180 | return -ENOMEM; | |
1181 | } | |
6a026589 | 1182 | |
83701246 YZ |
1183 | statret = __ceph_do_getattr(inode, page, |
1184 | CEPH_STAT_CAP_INLINE_DATA, !!page); | |
1185 | if (statret < 0) { | |
1186 | __free_page(page); | |
1187 | if (statret == -ENODATA) { | |
1188 | BUG_ON(retry_op != READ_INLINE); | |
1189 | goto again; | |
1190 | } | |
1191 | return statret; | |
1192 | } | |
6a026589 | 1193 | |
83701246 YZ |
1194 | i_size = i_size_read(inode); |
1195 | if (retry_op == READ_INLINE) { | |
fcc02d2a YZ |
1196 | BUG_ON(ret > 0 || read > 0); |
1197 | if (iocb->ki_pos < i_size && | |
09cbfeaf | 1198 | iocb->ki_pos < PAGE_SIZE) { |
83701246 YZ |
1199 | loff_t end = min_t(loff_t, i_size, |
1200 | iocb->ki_pos + len); | |
09cbfeaf | 1201 | end = min_t(loff_t, end, PAGE_SIZE); |
83701246 YZ |
1202 | if (statret < end) |
1203 | zero_user_segment(page, statret, end); | |
1204 | ret = copy_page_to_iter(page, | |
1205 | iocb->ki_pos & ~PAGE_MASK, | |
1206 | end - iocb->ki_pos, to); | |
1207 | iocb->ki_pos += ret; | |
fcc02d2a YZ |
1208 | read += ret; |
1209 | } | |
1210 | if (iocb->ki_pos < i_size && read < len) { | |
1211 | size_t zlen = min_t(size_t, len - read, | |
1212 | i_size - iocb->ki_pos); | |
1213 | ret = iov_iter_zero(zlen, to); | |
1214 | iocb->ki_pos += ret; | |
1215 | read += ret; | |
83701246 YZ |
1216 | } |
1217 | __free_pages(page, 0); | |
fcc02d2a | 1218 | return read; |
83701246 | 1219 | } |
6a026589 SW |
1220 | |
1221 | /* hit EOF or hole? */ | |
83701246 | 1222 | if (retry_op == CHECK_EOF && iocb->ki_pos < i_size && |
fcc02d2a | 1223 | ret < len) { |
8eb4efb0 | 1224 | dout("sync_read hit hole, ppos %lld < size %lld" |
99c88e69 | 1225 | ", reading more\n", iocb->ki_pos, i_size); |
8eb4efb0 | 1226 | |
6a026589 | 1227 | read += ret; |
6a026589 | 1228 | len -= ret; |
c8fe9b17 | 1229 | retry_op = HAVE_RETRIED; |
6a026589 SW |
1230 | goto again; |
1231 | } | |
1232 | } | |
8eb4efb0 | 1233 | |
6a026589 SW |
1234 | if (ret >= 0) |
1235 | ret += read; | |
1236 | ||
124e68e7 SW |
1237 | return ret; |
1238 | } | |
1239 | ||
1240 | /* | |
1241 | * Take cap references to avoid releasing caps to MDS mid-write. | |
1242 | * | |
1243 | * If we are synchronous, and write with an old snap context, the OSD | |
1244 | * may return EOLDSNAPC. In that case, retry the write.. _after_ | |
1245 | * dropping our cap refs and allowing the pending snap to logically | |
1246 | * complete _before_ this write occurs. | |
1247 | * | |
1248 | * If we are near ENOSPC, write synchronously. | |
1249 | */ | |
4908b822 | 1250 | static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from) |
124e68e7 SW |
1251 | { |
1252 | struct file *file = iocb->ki_filp; | |
33caad32 | 1253 | struct ceph_file_info *fi = file->private_data; |
496ad9aa | 1254 | struct inode *inode = file_inode(file); |
124e68e7 | 1255 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 YS |
1256 | struct ceph_osd_client *osdc = |
1257 | &ceph_sb_to_client(inode->i_sb)->client->osdc; | |
f66fd9f0 | 1258 | struct ceph_cap_flush *prealloc_cf; |
3309dd04 | 1259 | ssize_t count, written = 0; |
03d254ed | 1260 | int err, want, got; |
3309dd04 | 1261 | loff_t pos; |
124e68e7 SW |
1262 | |
1263 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
1264 | return -EROFS; | |
1265 | ||
f66fd9f0 YZ |
1266 | prealloc_cf = ceph_alloc_cap_flush(); |
1267 | if (!prealloc_cf) | |
1268 | return -ENOMEM; | |
1269 | ||
5955102c | 1270 | inode_lock(inode); |
03d254ed | 1271 | |
03d254ed | 1272 | /* We can write back this queue in page reclaim */ |
de1414a6 | 1273 | current->backing_dev_info = inode_to_bdi(inode); |
03d254ed | 1274 | |
55b0b31c YZ |
1275 | if (iocb->ki_flags & IOCB_APPEND) { |
1276 | err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); | |
1277 | if (err < 0) | |
1278 | goto out; | |
1279 | } | |
1280 | ||
3309dd04 AV |
1281 | err = generic_write_checks(iocb, from); |
1282 | if (err <= 0) | |
03d254ed YZ |
1283 | goto out; |
1284 | ||
3309dd04 AV |
1285 | pos = iocb->ki_pos; |
1286 | count = iov_iter_count(from); | |
5fa8e0a1 | 1287 | err = file_remove_privs(file); |
03d254ed YZ |
1288 | if (err) |
1289 | goto out; | |
1290 | ||
1291 | err = file_update_time(file); | |
1292 | if (err) | |
1293 | goto out; | |
1294 | ||
28127bdd YZ |
1295 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
1296 | err = ceph_uninline_data(file, NULL); | |
1297 | if (err < 0) | |
1298 | goto out; | |
1299 | } | |
1300 | ||
124e68e7 | 1301 | retry_snap: |
6070e0c1 | 1302 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) { |
03d254ed | 1303 | err = -ENOSPC; |
6070e0c1 YZ |
1304 | goto out; |
1305 | } | |
03d254ed | 1306 | |
ac7f29bf | 1307 | dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n", |
99c88e69 | 1308 | inode, ceph_vinop(inode), pos, count, i_size_read(inode)); |
7971bd92 SW |
1309 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
1310 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1311 | else | |
1312 | want = CEPH_CAP_FILE_BUFFER; | |
03d254ed | 1313 | got = 0; |
3738daa6 YZ |
1314 | err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count, |
1315 | &got, NULL); | |
03d254ed | 1316 | if (err < 0) |
37505d57 | 1317 | goto out; |
124e68e7 | 1318 | |
ac7f29bf | 1319 | dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n", |
03d254ed | 1320 | inode, ceph_vinop(inode), pos, count, ceph_cap_string(got)); |
7971bd92 SW |
1321 | |
1322 | if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || | |
2ba48ce5 | 1323 | (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC)) { |
5dda377c | 1324 | struct ceph_snap_context *snapc; |
4908b822 | 1325 | struct iov_iter data; |
5955102c | 1326 | inode_unlock(inode); |
5dda377c YZ |
1327 | |
1328 | spin_lock(&ci->i_ceph_lock); | |
1329 | if (__ceph_have_pending_cap_snap(ci)) { | |
1330 | struct ceph_cap_snap *capsnap = | |
1331 | list_last_entry(&ci->i_cap_snaps, | |
1332 | struct ceph_cap_snap, | |
1333 | ci_item); | |
1334 | snapc = ceph_get_snap_context(capsnap->context); | |
1335 | } else { | |
1336 | BUG_ON(!ci->i_head_snapc); | |
1337 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
1338 | } | |
1339 | spin_unlock(&ci->i_ceph_lock); | |
1340 | ||
4908b822 AV |
1341 | /* we might need to revert back to that point */ |
1342 | data = *from; | |
2ba48ce5 | 1343 | if (iocb->ki_flags & IOCB_DIRECT) |
c8fe9b17 YZ |
1344 | written = ceph_direct_read_write(iocb, &data, snapc, |
1345 | &prealloc_cf); | |
e8344e66 | 1346 | else |
5dda377c | 1347 | written = ceph_sync_write(iocb, &data, pos, snapc); |
0e5dd45c | 1348 | if (written == -EOLDSNAPC) { |
1349 | dout("aio_write %p %llx.%llx %llu~%u" | |
1350 | "got EOLDSNAPC, retrying\n", | |
1351 | inode, ceph_vinop(inode), | |
4908b822 | 1352 | pos, (unsigned)count); |
5955102c | 1353 | inode_lock(inode); |
0e5dd45c | 1354 | goto retry_snap; |
1355 | } | |
4908b822 AV |
1356 | if (written > 0) |
1357 | iov_iter_advance(from, written); | |
5dda377c | 1358 | ceph_put_snap_context(snapc); |
7971bd92 | 1359 | } else { |
99c88e69 | 1360 | loff_t old_size = i_size_read(inode); |
b0d7c223 YZ |
1361 | /* |
1362 | * No need to acquire the i_truncate_mutex. Because | |
1363 | * the MDS revokes Fwb caps before sending truncate | |
1364 | * message to us. We can't get Fwb cap while there | |
1365 | * are pending vmtruncate. So write and vmtruncate | |
1366 | * can not run at the same time | |
1367 | */ | |
4908b822 | 1368 | written = generic_perform_write(file, from, pos); |
aec605f4 AV |
1369 | if (likely(written >= 0)) |
1370 | iocb->ki_pos = pos + written; | |
99c88e69 | 1371 | if (i_size_read(inode) > old_size) |
32d3e148 | 1372 | ceph_fscache_update_objectsize(inode); |
5955102c | 1373 | inode_unlock(inode); |
7971bd92 | 1374 | } |
d8de9ab6 | 1375 | |
03d254ed | 1376 | if (written >= 0) { |
fca65b4a | 1377 | int dirty; |
be655596 | 1378 | spin_lock(&ci->i_ceph_lock); |
28127bdd | 1379 | ci->i_inline_version = CEPH_INLINE_NONE; |
f66fd9f0 YZ |
1380 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
1381 | &prealloc_cf); | |
be655596 | 1382 | spin_unlock(&ci->i_ceph_lock); |
fca65b4a SW |
1383 | if (dirty) |
1384 | __mark_inode_dirty(inode, dirty); | |
124e68e7 | 1385 | } |
7971bd92 | 1386 | |
124e68e7 | 1387 | dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n", |
4908b822 | 1388 | inode, ceph_vinop(inode), pos, (unsigned)count, |
7971bd92 | 1389 | ceph_cap_string(got)); |
124e68e7 | 1390 | ceph_put_cap_refs(ci, got); |
7971bd92 | 1391 | |
03d254ed | 1392 | if (written >= 0 && |
6070e0c1 YZ |
1393 | ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) || |
1394 | ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) { | |
03d254ed | 1395 | err = vfs_fsync_range(file, pos, pos + written - 1, 1); |
6070e0c1 | 1396 | if (err < 0) |
03d254ed | 1397 | written = err; |
6070e0c1 | 1398 | } |
03d254ed | 1399 | |
2f75e9e1 SW |
1400 | goto out_unlocked; |
1401 | ||
03d254ed | 1402 | out: |
5955102c | 1403 | inode_unlock(inode); |
2f75e9e1 | 1404 | out_unlocked: |
f66fd9f0 | 1405 | ceph_free_cap_flush(prealloc_cf); |
03d254ed | 1406 | current->backing_dev_info = NULL; |
03d254ed | 1407 | return written ? written : err; |
124e68e7 SW |
1408 | } |
1409 | ||
1410 | /* | |
1411 | * llseek. be sure to verify file size on SEEK_END. | |
1412 | */ | |
965c8e59 | 1413 | static loff_t ceph_llseek(struct file *file, loff_t offset, int whence) |
124e68e7 SW |
1414 | { |
1415 | struct inode *inode = file->f_mapping->host; | |
99c88e69 | 1416 | loff_t i_size; |
124e68e7 SW |
1417 | int ret; |
1418 | ||
5955102c | 1419 | inode_lock(inode); |
6a82c47a | 1420 | |
965c8e59 | 1421 | if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { |
508b32d8 | 1422 | ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); |
124e68e7 SW |
1423 | if (ret < 0) { |
1424 | offset = ret; | |
1425 | goto out; | |
1426 | } | |
06222e49 JB |
1427 | } |
1428 | ||
99c88e69 | 1429 | i_size = i_size_read(inode); |
965c8e59 | 1430 | switch (whence) { |
06222e49 | 1431 | case SEEK_END: |
99c88e69 | 1432 | offset += i_size; |
124e68e7 SW |
1433 | break; |
1434 | case SEEK_CUR: | |
1435 | /* | |
1436 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
1437 | * position-querying operation. Avoid rewriting the "same" | |
1438 | * f_pos value back to the file because a concurrent read(), | |
1439 | * write() or lseek() might have altered it | |
1440 | */ | |
1441 | if (offset == 0) { | |
1442 | offset = file->f_pos; | |
1443 | goto out; | |
1444 | } | |
1445 | offset += file->f_pos; | |
1446 | break; | |
06222e49 | 1447 | case SEEK_DATA: |
99c88e69 | 1448 | if (offset >= i_size) { |
06222e49 JB |
1449 | ret = -ENXIO; |
1450 | goto out; | |
1451 | } | |
1452 | break; | |
1453 | case SEEK_HOLE: | |
99c88e69 | 1454 | if (offset >= i_size) { |
06222e49 JB |
1455 | ret = -ENXIO; |
1456 | goto out; | |
1457 | } | |
99c88e69 | 1458 | offset = i_size; |
06222e49 | 1459 | break; |
124e68e7 SW |
1460 | } |
1461 | ||
46a1c2c7 | 1462 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
124e68e7 SW |
1463 | |
1464 | out: | |
5955102c | 1465 | inode_unlock(inode); |
124e68e7 SW |
1466 | return offset; |
1467 | } | |
1468 | ||
ad7a60de LW |
1469 | static inline void ceph_zero_partial_page( |
1470 | struct inode *inode, loff_t offset, unsigned size) | |
1471 | { | |
1472 | struct page *page; | |
09cbfeaf | 1473 | pgoff_t index = offset >> PAGE_SHIFT; |
ad7a60de LW |
1474 | |
1475 | page = find_lock_page(inode->i_mapping, index); | |
1476 | if (page) { | |
1477 | wait_on_page_writeback(page); | |
09cbfeaf | 1478 | zero_user(page, offset & (PAGE_SIZE - 1), size); |
ad7a60de | 1479 | unlock_page(page); |
09cbfeaf | 1480 | put_page(page); |
ad7a60de LW |
1481 | } |
1482 | } | |
1483 | ||
1484 | static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset, | |
1485 | loff_t length) | |
1486 | { | |
09cbfeaf | 1487 | loff_t nearly = round_up(offset, PAGE_SIZE); |
ad7a60de LW |
1488 | if (offset < nearly) { |
1489 | loff_t size = nearly - offset; | |
1490 | if (length < size) | |
1491 | size = length; | |
1492 | ceph_zero_partial_page(inode, offset, size); | |
1493 | offset += size; | |
1494 | length -= size; | |
1495 | } | |
09cbfeaf KS |
1496 | if (length >= PAGE_SIZE) { |
1497 | loff_t size = round_down(length, PAGE_SIZE); | |
ad7a60de LW |
1498 | truncate_pagecache_range(inode, offset, offset + size - 1); |
1499 | offset += size; | |
1500 | length -= size; | |
1501 | } | |
1502 | if (length) | |
1503 | ceph_zero_partial_page(inode, offset, length); | |
1504 | } | |
1505 | ||
1506 | static int ceph_zero_partial_object(struct inode *inode, | |
1507 | loff_t offset, loff_t *length) | |
1508 | { | |
1509 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1510 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
1511 | struct ceph_osd_request *req; | |
1512 | int ret = 0; | |
1513 | loff_t zero = 0; | |
1514 | int op; | |
1515 | ||
1516 | if (!length) { | |
1517 | op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE; | |
1518 | length = &zero; | |
1519 | } else { | |
1520 | op = CEPH_OSD_OP_ZERO; | |
1521 | } | |
1522 | ||
1523 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1524 | ceph_vino(inode), | |
1525 | offset, length, | |
715e4cd4 | 1526 | 0, 1, op, |
ad7a60de LW |
1527 | CEPH_OSD_FLAG_WRITE | |
1528 | CEPH_OSD_FLAG_ONDISK, | |
1529 | NULL, 0, 0, false); | |
1530 | if (IS_ERR(req)) { | |
1531 | ret = PTR_ERR(req); | |
1532 | goto out; | |
1533 | } | |
1534 | ||
1535 | ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap, | |
1536 | &inode->i_mtime); | |
1537 | ||
1538 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); | |
1539 | if (!ret) { | |
1540 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1541 | if (ret == -ENOENT) | |
1542 | ret = 0; | |
1543 | } | |
1544 | ceph_osdc_put_request(req); | |
1545 | ||
1546 | out: | |
1547 | return ret; | |
1548 | } | |
1549 | ||
1550 | static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length) | |
1551 | { | |
1552 | int ret = 0; | |
1553 | struct ceph_inode_info *ci = ceph_inode(inode); | |
b314a90d SW |
1554 | s32 stripe_unit = ceph_file_layout_su(ci->i_layout); |
1555 | s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout); | |
1556 | s32 object_size = ceph_file_layout_object_size(ci->i_layout); | |
1557 | u64 object_set_size = object_size * stripe_count; | |
1558 | u64 nearly, t; | |
1559 | ||
1560 | /* round offset up to next period boundary */ | |
1561 | nearly = offset + object_set_size - 1; | |
1562 | t = nearly; | |
1563 | nearly -= do_div(t, object_set_size); | |
ad7a60de | 1564 | |
ad7a60de LW |
1565 | while (length && offset < nearly) { |
1566 | loff_t size = length; | |
1567 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1568 | if (ret < 0) | |
1569 | return ret; | |
1570 | offset += size; | |
1571 | length -= size; | |
1572 | } | |
1573 | while (length >= object_set_size) { | |
1574 | int i; | |
1575 | loff_t pos = offset; | |
1576 | for (i = 0; i < stripe_count; ++i) { | |
1577 | ret = ceph_zero_partial_object(inode, pos, NULL); | |
1578 | if (ret < 0) | |
1579 | return ret; | |
1580 | pos += stripe_unit; | |
1581 | } | |
1582 | offset += object_set_size; | |
1583 | length -= object_set_size; | |
1584 | } | |
1585 | while (length) { | |
1586 | loff_t size = length; | |
1587 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1588 | if (ret < 0) | |
1589 | return ret; | |
1590 | offset += size; | |
1591 | length -= size; | |
1592 | } | |
1593 | return ret; | |
1594 | } | |
1595 | ||
1596 | static long ceph_fallocate(struct file *file, int mode, | |
1597 | loff_t offset, loff_t length) | |
1598 | { | |
1599 | struct ceph_file_info *fi = file->private_data; | |
aa8b60e0 | 1600 | struct inode *inode = file_inode(file); |
ad7a60de LW |
1601 | struct ceph_inode_info *ci = ceph_inode(inode); |
1602 | struct ceph_osd_client *osdc = | |
1603 | &ceph_inode_to_client(inode)->client->osdc; | |
f66fd9f0 | 1604 | struct ceph_cap_flush *prealloc_cf; |
ad7a60de LW |
1605 | int want, got = 0; |
1606 | int dirty; | |
1607 | int ret = 0; | |
1608 | loff_t endoff = 0; | |
1609 | loff_t size; | |
1610 | ||
494d77bf YZ |
1611 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
1612 | return -EOPNOTSUPP; | |
1613 | ||
ad7a60de LW |
1614 | if (!S_ISREG(inode->i_mode)) |
1615 | return -EOPNOTSUPP; | |
1616 | ||
f66fd9f0 YZ |
1617 | prealloc_cf = ceph_alloc_cap_flush(); |
1618 | if (!prealloc_cf) | |
1619 | return -ENOMEM; | |
1620 | ||
5955102c | 1621 | inode_lock(inode); |
ad7a60de LW |
1622 | |
1623 | if (ceph_snap(inode) != CEPH_NOSNAP) { | |
1624 | ret = -EROFS; | |
1625 | goto unlock; | |
1626 | } | |
1627 | ||
1628 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) && | |
1629 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
1630 | ret = -ENOSPC; | |
1631 | goto unlock; | |
1632 | } | |
1633 | ||
28127bdd YZ |
1634 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
1635 | ret = ceph_uninline_data(file, NULL); | |
1636 | if (ret < 0) | |
1637 | goto unlock; | |
1638 | } | |
1639 | ||
ad7a60de LW |
1640 | size = i_size_read(inode); |
1641 | if (!(mode & FALLOC_FL_KEEP_SIZE)) | |
1642 | endoff = offset + length; | |
1643 | ||
1644 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1645 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1646 | else | |
1647 | want = CEPH_CAP_FILE_BUFFER; | |
1648 | ||
3738daa6 | 1649 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL); |
ad7a60de LW |
1650 | if (ret < 0) |
1651 | goto unlock; | |
1652 | ||
1653 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1654 | if (offset < size) | |
1655 | ceph_zero_pagecache_range(inode, offset, length); | |
1656 | ret = ceph_zero_objects(inode, offset, length); | |
1657 | } else if (endoff > size) { | |
1658 | truncate_pagecache_range(inode, size, -1); | |
1659 | if (ceph_inode_set_size(inode, endoff)) | |
1660 | ceph_check_caps(ceph_inode(inode), | |
1661 | CHECK_CAPS_AUTHONLY, NULL); | |
1662 | } | |
1663 | ||
1664 | if (!ret) { | |
1665 | spin_lock(&ci->i_ceph_lock); | |
28127bdd | 1666 | ci->i_inline_version = CEPH_INLINE_NONE; |
f66fd9f0 YZ |
1667 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
1668 | &prealloc_cf); | |
ad7a60de LW |
1669 | spin_unlock(&ci->i_ceph_lock); |
1670 | if (dirty) | |
1671 | __mark_inode_dirty(inode, dirty); | |
1672 | } | |
1673 | ||
1674 | ceph_put_cap_refs(ci, got); | |
1675 | unlock: | |
5955102c | 1676 | inode_unlock(inode); |
f66fd9f0 | 1677 | ceph_free_cap_flush(prealloc_cf); |
ad7a60de LW |
1678 | return ret; |
1679 | } | |
1680 | ||
124e68e7 SW |
1681 | const struct file_operations ceph_file_fops = { |
1682 | .open = ceph_open, | |
1683 | .release = ceph_release, | |
1684 | .llseek = ceph_llseek, | |
3644424d | 1685 | .read_iter = ceph_read_iter, |
4908b822 | 1686 | .write_iter = ceph_write_iter, |
124e68e7 SW |
1687 | .mmap = ceph_mmap, |
1688 | .fsync = ceph_fsync, | |
40819f6f GF |
1689 | .lock = ceph_lock, |
1690 | .flock = ceph_flock, | |
124e68e7 | 1691 | .splice_read = generic_file_splice_read, |
3551dd79 | 1692 | .splice_write = iter_file_splice_write, |
124e68e7 SW |
1693 | .unlocked_ioctl = ceph_ioctl, |
1694 | .compat_ioctl = ceph_ioctl, | |
ad7a60de | 1695 | .fallocate = ceph_fallocate, |
124e68e7 SW |
1696 | }; |
1697 |