blob: 3c68e6aee2f0531bdd8afb0125bbae48419d0dfb [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weil124e68e2009-10-06 11:31:08 -07002
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003#include <linux/module.h>
Sage Weil124e68e2009-10-06 11:31:08 -07004#include <linux/sched.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09005#include <linux/slab.h>
Sage Weil124e68e2009-10-06 11:31:08 -07006#include <linux/file.h>
Sage Weil5ef50c32012-07-31 11:27:36 -07007#include <linux/mount.h>
Sage Weil124e68e2009-10-06 11:31:08 -07008#include <linux/namei.h>
9#include <linux/writeback.h>
Li Wangad7a60de2013-08-15 11:51:44 +080010#include <linux/falloc.h>
Sage Weil124e68e2009-10-06 11:31:08 -070011
12#include "super.h"
13#include "mds_client.h"
Milosz Tanski99ccbd22013-08-21 17:29:54 -040014#include "cache.h"
Sage Weil124e68e2009-10-06 11:31:08 -070015
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
Zhu, Caifengb5b98989d2015-10-08 15:26:15 +080037/*
38 * Calculate the length sum of direct io vectors that can
39 * be combined into one page vector.
40 */
41static 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 */
65static struct page **
66dio_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;
101fail:
102 ceph_put_page_vector(pages, idx, false);
103 return ERR_PTR(ret);
104}
Sage Weil124e68e2009-10-06 11:31:08 -0700105
106/*
107 * Prepare an open request. Preallocate ceph_cap to avoid an
108 * inopportune ENOMEM later.
109 */
110static struct ceph_mds_request *
111prepare_open_request(struct super_block *sb, int flags, int create_mode)
112{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700113 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
114 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700115 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);
Sage Weil124e68e2009-10-06 11:31:08 -0700128out:
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 */
136static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
137{
138 struct ceph_file_info *cf;
139 int ret = 0;
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400140 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;
Sage Weil124e68e2009-10-06 11:31:08 -0700143
144 switch (inode->i_mode & S_IFMT) {
145 case S_IFREG:
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400146 /* 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);
Sage Weil124e68e2009-10-06 11:31:08 -0700157 case S_IFDIR:
158 dout("init_file %p %p 0%o (regular)\n", inode, file,
159 inode->i_mode);
Yan, Zheng687265e2015-06-13 17:27:05 +0800160 cf = kmem_cache_alloc(ceph_file_cachep, GFP_KERNEL | __GFP_ZERO);
Sage Weil124e68e2009-10-06 11:31:08 -0700161 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;
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800167 cf->readdir_cache_idx = -1;
Sage Weil124e68e2009-10-06 11:31:08 -0700168 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/*
Sage Weil124e68e2009-10-06 11:31:08 -0700195 * 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 */
200int ceph_open(struct inode *inode, struct file *file)
201{
202 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700203 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
204 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700205 struct ceph_mds_request *req;
206 struct ceph_file_info *cf = file->private_data;
Sage Weil124e68e2009-10-06 11:31:08 -0700207 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) {
Sage Weilbe655592011-11-30 09:47:09 -0800231 spin_lock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700232 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800233 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700234 return ceph_init_file(inode, file, fmode);
235 }
236
237 /*
Sage Weil7421ab82010-11-07 09:07:15 -0800238 * No need to block if we have caps on the auth MDS (for
239 * write) or any MDS (for read). Update wanted set
Sage Weil124e68e2009-10-06 11:31:08 -0700240 * asynchronously.
241 */
Sage Weilbe655592011-11-30 09:47:09 -0800242 spin_lock(&ci->i_ceph_lock);
Sage Weil7421ab82010-11-07 09:07:15 -0800243 if (__ceph_is_any_real_caps(ci) &&
244 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
Sage Weil124e68e2009-10-06 11:31:08 -0700245 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);
Sage Weilbe655592011-11-30 09:47:09 -0800252 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700253
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);
Sage Weilbe655592011-11-30 09:47:09 -0800264 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700265 return ceph_init_file(inode, file, fmode);
266 }
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400267
Sage Weilbe655592011-11-30 09:47:09 -0800268 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700269
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 }
Sage Weil70b666c2011-05-27 09:24:26 -0700276 req->r_inode = inode;
277 ihold(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400278
Sage Weil124e68e2009-10-06 11:31:08 -0700279 req->r_num_caps = 1;
Jianpeng Mae36d5712015-08-18 10:25:35 +0800280 err = ceph_mdsc_do_request(mdsc, NULL, req);
Sage Weil124e68e2009-10-06 11:31:08 -0700281 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));
285out:
286 return err;
287}
288
289
290/*
Sage Weil5ef50c32012-07-31 11:27:36 -0700291 * 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.
Sage Weil124e68e2009-10-06 11:31:08 -0700293 */
Sage Weil5ef50c32012-07-31 11:27:36 -0700294int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
Al Viro30d904942012-06-22 12:40:19 +0400295 struct file *file, unsigned flags, umode_t mode,
Al Virod9585272012-06-22 12:39:14 +0400296 int *opened)
Sage Weil124e68e2009-10-06 11:31:08 -0700297{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700298 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
299 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700300 struct ceph_mds_request *req;
Sage Weil5ef50c32012-07-31 11:27:36 -0700301 struct dentry *dn;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800302 struct ceph_acls_info acls = {};
Sage Weil124e68e2009-10-06 11:31:08 -0700303 int err;
Sage Weil124e68e2009-10-06 11:31:08 -0700304
Al Viroa4555892014-10-21 20:11:25 -0400305 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
306 dir, dentry, dentry,
Sage Weil5ef50c32012-07-31 11:27:36 -0700307 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
308
309 if (dentry->d_name.len > NAME_MAX)
310 return -ENAMETOOLONG;
311
312 err = ceph_init_dentry(dentry);
313 if (err < 0)
314 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700315
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800316 if (flags & O_CREAT) {
317 err = ceph_pre_init_acls(dir, &mode, &acls);
318 if (err < 0)
319 return err;
320 }
321
Sage Weil124e68e2009-10-06 11:31:08 -0700322 /* do the open */
323 req = prepare_open_request(dir->i_sb, flags, mode);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800324 if (IS_ERR(req)) {
325 err = PTR_ERR(req);
326 goto out_acl;
327 }
Sage Weil124e68e2009-10-06 11:31:08 -0700328 req->r_dentry = dget(dentry);
329 req->r_num_caps = 2;
330 if (flags & O_CREAT) {
331 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
332 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800333 if (acls.pagelist) {
334 req->r_pagelist = acls.pagelist;
335 acls.pagelist = NULL;
336 }
Sage Weil124e68e2009-10-06 11:31:08 -0700337 }
338 req->r_locked_dir = dir; /* caller holds dir->i_mutex */
Sage Weilacda7652011-07-26 11:27:48 -0700339 err = ceph_mdsc_do_request(mdsc,
340 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
341 req);
Yan, Zhengbf91c312015-01-19 13:23:20 +0800342 err = ceph_handle_snapdir(req, dentry, err);
Sam Lang79aec9842012-12-19 09:44:23 -1000343 if (err)
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800344 goto out_req;
Sam Lang79aec9842012-12-19 09:44:23 -1000345
Jianpeng Maa43137f2015-08-18 10:23:50 +0800346 if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
Sage Weil124e68e2009-10-06 11:31:08 -0700347 err = ceph_handle_notrace_create(dir, dentry);
Sage Weil5ef50c32012-07-31 11:27:36 -0700348
349 if (d_unhashed(dentry)) {
350 dn = ceph_finish_lookup(req, dentry, err);
351 if (IS_ERR(dn))
352 err = PTR_ERR(dn);
353 } else {
354 /* we were given a hashed negative dentry */
355 dn = NULL;
356 }
Sage Weil468640e2011-07-26 11:28:11 -0700357 if (err)
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800358 goto out_req;
David Howells2b0143b2015-03-17 22:25:59 +0000359 if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) {
Sage Weil5ef50c32012-07-31 11:27:36 -0700360 /* make vfs retry on splice, ENOENT, or symlink */
361 dout("atomic_open finish_no_open on dn %p\n", dn);
362 err = finish_no_open(file, dn);
363 } else {
364 dout("atomic_open finish_open on dn %p\n", dn);
Sam Lang6e8575f2012-12-28 09:56:46 -0800365 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
David Howells2b0143b2015-03-17 22:25:59 +0000366 ceph_init_inode_acls(d_inode(dentry), &acls);
Sam Lang6e8575f2012-12-28 09:56:46 -0800367 *opened |= FILE_CREATED;
368 }
Sage Weil5ef50c32012-07-31 11:27:36 -0700369 err = finish_open(file, dentry, ceph_open, opened);
370 }
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800371out_req:
Yan, Zhengab866542014-04-01 20:20:17 +0800372 if (!req->r_err && req->r_target_inode)
373 ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode);
Sage Weil124e68e2009-10-06 11:31:08 -0700374 ceph_mdsc_put_request(req);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800375out_acl:
376 ceph_release_acls_info(&acls);
Sage Weil5ef50c32012-07-31 11:27:36 -0700377 dout("atomic_open result=%d\n", err);
Al Virod9585272012-06-22 12:39:14 +0400378 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700379}
380
381int ceph_release(struct inode *inode, struct file *file)
382{
383 struct ceph_inode_info *ci = ceph_inode(inode);
384 struct ceph_file_info *cf = file->private_data;
385
386 dout("release inode %p file %p\n", inode, file);
387 ceph_put_fmode(ci, cf->fmode);
388 if (cf->last_readdir)
389 ceph_mdsc_put_request(cf->last_readdir);
390 kfree(cf->last_name);
391 kfree(cf->dir_info);
Sage Weil124e68e2009-10-06 11:31:08 -0700392 kmem_cache_free(ceph_file_cachep, cf);
Sage Weil195d3ce2010-03-01 09:57:54 -0800393
394 /* wake up anyone waiting for caps on this inode */
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700395 wake_up_all(&ci->i_cap_wq);
Sage Weil124e68e2009-10-06 11:31:08 -0700396 return 0;
397}
398
Yan, Zheng83701242014-11-14 22:36:18 +0800399enum {
400 CHECK_EOF = 1,
401 READ_INLINE = 2,
402};
403
Sage Weil124e68e2009-10-06 11:31:08 -0700404/*
Sage Weil124e68e2009-10-06 11:31:08 -0700405 * Read a range of bytes striped over one or more objects. Iterate over
406 * objects we stripe over. (That's not atomic, but good enough for now.)
407 *
408 * If we get a short result from the OSD, check against i_size; we need to
409 * only return a short read to the caller if we hit EOF.
410 */
411static int striped_read(struct inode *inode,
412 u64 off, u64 len,
Sage Weil6a026582010-02-09 14:04:02 -0800413 struct page **pages, int num_pages,
Sage Weilc3cd6282011-06-01 16:08:44 -0700414 int *checkeof, bool o_direct,
Henry C Changab226e22010-12-15 20:41:54 -0800415 unsigned long buf_align)
Sage Weil124e68e2009-10-06 11:31:08 -0700416{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700417 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700418 struct ceph_inode_info *ci = ceph_inode(inode);
Dan Carpenter688bac42013-07-23 16:48:01 +0300419 u64 pos, this_len, left;
Sage Weilb7495fc2010-11-09 12:43:12 -0800420 int io_align, page_align;
Dan Carpenter688bac42013-07-23 16:48:01 +0300421 int pages_left;
Sage Weil124e68e2009-10-06 11:31:08 -0700422 int read;
423 struct page **page_pos;
424 int ret;
425 bool hit_stripe, was_short;
426
427 /*
428 * we may need to do multiple reads. not atomic, unfortunately.
429 */
430 pos = off;
431 left = len;
432 page_pos = pages;
433 pages_left = num_pages;
434 read = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -0800435 io_align = off & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700436
437more:
Sage Weilc3cd6282011-06-01 16:08:44 -0700438 if (o_direct)
Henry C Changab226e22010-12-15 20:41:54 -0800439 page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
Sage Weilb7495fc2010-11-09 12:43:12 -0800440 else
441 page_align = pos & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700442 this_len = left;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700443 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
Sage Weil124e68e2009-10-06 11:31:08 -0700444 &ci->i_layout, pos, &this_len,
445 ci->i_truncate_seq,
446 ci->i_truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -0800447 page_pos, pages_left, page_align);
Sage Weil124e68e2009-10-06 11:31:08 -0700448 if (ret == -ENOENT)
449 ret = 0;
Sage Weil0e987282011-06-07 20:40:35 -0700450 hit_stripe = this_len < left;
451 was_short = ret >= 0 && ret < this_len;
Dan Carpenter688bac42013-07-23 16:48:01 +0300452 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read,
Sage Weil124e68e2009-10-06 11:31:08 -0700453 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
454
majianpeng02ae66d2013-08-06 16:20:38 +0800455 if (ret >= 0) {
456 int didpages;
457 if (was_short && (pos + ret < inode->i_size)) {
Yan, Zheng1487a682015-01-06 15:29:14 +0800458 int zlen = min(this_len - ret,
459 inode->i_size - pos - ret);
460 int zoff = (o_direct ? buf_align : io_align) +
461 read + ret;
majianpeng02ae66d2013-08-06 16:20:38 +0800462 dout(" zero gap %llu to %llu\n",
Yan, Zheng1487a682015-01-06 15:29:14 +0800463 pos + ret, pos + ret + zlen);
464 ceph_zero_page_vector_range(zoff, zlen, pages);
465 ret += zlen;
Sage Weil124e68e2009-10-06 11:31:08 -0700466 }
majianpeng02ae66d2013-08-06 16:20:38 +0800467
468 didpages = (page_align + ret) >> PAGE_CACHE_SHIFT;
Sage Weil124e68e2009-10-06 11:31:08 -0700469 pos += ret;
470 read = pos - off;
471 left -= ret;
472 page_pos += didpages;
473 pages_left -= didpages;
474
majianpeng02ae66d2013-08-06 16:20:38 +0800475 /* hit stripe and need continue*/
476 if (left && hit_stripe && pos < inode->i_size)
Sage Weil124e68e2009-10-06 11:31:08 -0700477 goto more;
478 }
479
majianpengee7289b2013-08-21 15:02:51 +0800480 if (read > 0) {
majianpeng02ae66d2013-08-06 16:20:38 +0800481 ret = read;
Sage Weilc3cd6282011-06-01 16:08:44 -0700482 /* did we bounce off eof? */
483 if (pos + left > inode->i_size)
Yan, Zheng83701242014-11-14 22:36:18 +0800484 *checkeof = CHECK_EOF;
Sage Weil124e68e2009-10-06 11:31:08 -0700485 }
486
Sage Weil124e68e2009-10-06 11:31:08 -0700487 dout("striped_read returns %d\n", ret);
488 return ret;
489}
490
491/*
492 * Completely synchronous read and write methods. Direct from __user
493 * buffer to osd, or directly to user pages (if O_DIRECT).
494 *
495 * If the read spans object boundary, just do multiple reads.
496 */
majianpeng8eb4efb2013-09-26 14:42:17 +0800497static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i,
498 int *checkeof)
Sage Weil124e68e2009-10-06 11:31:08 -0700499{
majianpeng8eb4efb2013-09-26 14:42:17 +0800500 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -0500501 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700502 struct page **pages;
majianpeng8eb4efb2013-09-26 14:42:17 +0800503 u64 off = iocb->ki_pos;
Henry C Changab226e22010-12-15 20:41:54 -0800504 int num_pages, ret;
Al Viro2b777c92014-04-03 22:31:22 -0400505 size_t len = iov_iter_count(i);
Sage Weil124e68e2009-10-06 11:31:08 -0700506
majianpeng8eb4efb2013-09-26 14:42:17 +0800507 dout("sync_read on file %p %llu~%u %s\n", file, off,
508 (unsigned)len,
Sage Weil124e68e2009-10-06 11:31:08 -0700509 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
Yan, Zhengd0d0db22014-07-21 10:15:48 +0800510
511 if (!len)
512 return 0;
Sage Weile98b6fe2010-11-09 12:24:53 -0800513 /*
514 * flush any page cache pages in this range. this
515 * will make concurrent normal and sync io slow,
516 * but it will at least behave sensibly when they are
517 * in sequence.
518 */
majianpeng8eb4efb2013-09-26 14:42:17 +0800519 ret = filemap_write_and_wait_range(inode->i_mapping, off,
520 off + len);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800521 if (ret < 0)
majianpeng8eb4efb2013-09-26 14:42:17 +0800522 return ret;
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800523
Al Viro2ba48ce2015-04-09 13:52:01 -0400524 if (iocb->ki_flags & IOCB_DIRECT) {
majianpeng8eb4efb2013-09-26 14:42:17 +0800525 while (iov_iter_count(i)) {
Al Viro2b777c92014-04-03 22:31:22 -0400526 size_t start;
527 ssize_t n;
Sage Weil124e68e2009-10-06 11:31:08 -0700528
Zhu, Caifengb5b98989d2015-10-08 15:26:15 +0800529 n = dio_get_pagev_size(i);
530 pages = dio_get_pages_alloc(i, n, &start, &num_pages);
531 if (IS_ERR(pages))
532 return PTR_ERR(pages);
Al Viro2b777c92014-04-03 22:31:22 -0400533
534 ret = striped_read(inode, off, n,
majianpeng8eb4efb2013-09-26 14:42:17 +0800535 pages, num_pages, checkeof,
Al Viro2b777c92014-04-03 22:31:22 -0400536 1, start);
537
majianpeng8eb4efb2013-09-26 14:42:17 +0800538 ceph_put_page_vector(pages, num_pages, true);
539
540 if (ret <= 0)
541 break;
542 off += ret;
543 iov_iter_advance(i, ret);
Al Viro2b777c92014-04-03 22:31:22 -0400544 if (ret < n)
majianpeng8eb4efb2013-09-26 14:42:17 +0800545 break;
546 }
547 } else {
548 num_pages = calc_pages_for(off, len);
Yan, Zheng687265e2015-06-13 17:27:05 +0800549 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
majianpeng8eb4efb2013-09-26 14:42:17 +0800550 if (IS_ERR(pages))
551 return PTR_ERR(pages);
552 ret = striped_read(inode, off, len, pages,
553 num_pages, checkeof, 0, 0);
554 if (ret > 0) {
555 int l, k = 0;
Al Viro2b777c92014-04-03 22:31:22 -0400556 size_t left = ret;
majianpeng8eb4efb2013-09-26 14:42:17 +0800557
558 while (left) {
Yan, Zheng5aaa4322014-07-02 10:13:00 +0800559 size_t page_off = off & ~PAGE_MASK;
560 size_t copy = min_t(size_t,
561 PAGE_SIZE - page_off, left);
562 l = copy_page_to_iter(pages[k++], page_off,
563 copy, i);
Al Viro2b777c92014-04-03 22:31:22 -0400564 off += l;
565 left -= l;
566 if (l < copy)
majianpeng8eb4efb2013-09-26 14:42:17 +0800567 break;
568 }
569 }
Sage Weil124e68e2009-10-06 11:31:08 -0700570 ceph_release_page_vector(pages, num_pages);
majianpeng8eb4efb2013-09-26 14:42:17 +0800571 }
572
573 if (off > iocb->ki_pos) {
574 ret = off - iocb->ki_pos;
575 iocb->ki_pos = off;
576 }
577
Sage Weil124e68e2009-10-06 11:31:08 -0700578 dout("sync_read result %d\n", ret);
579 return ret;
580}
581
582/*
Alex Elder26be8802013-04-15 11:20:42 -0500583 * Write commit request unsafe callback, called to tell us when a
584 * request is unsafe (that is, in flight--has been handed to the
585 * messenger to send to its target osd). It is called again when
586 * we've received a response message indicating the request is
587 * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
588 * is completed early (and unsuccessfully) due to a timeout or
589 * interrupt.
590 *
591 * This is used if we requested both an ACK and ONDISK commit reply
592 * from the OSD.
Sage Weil124e68e2009-10-06 11:31:08 -0700593 */
Alex Elder26be8802013-04-15 11:20:42 -0500594static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe)
Sage Weil124e68e2009-10-06 11:31:08 -0700595{
596 struct ceph_inode_info *ci = ceph_inode(req->r_inode);
597
Alex Elder26be8802013-04-15 11:20:42 -0500598 dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid,
599 unsafe ? "un" : "");
600 if (unsafe) {
601 ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR);
602 spin_lock(&ci->i_unsafe_lock);
603 list_add_tail(&req->r_unsafe_item,
604 &ci->i_unsafe_writes);
605 spin_unlock(&ci->i_unsafe_lock);
606 } else {
607 spin_lock(&ci->i_unsafe_lock);
608 list_del_init(&req->r_unsafe_item);
609 spin_unlock(&ci->i_unsafe_lock);
610 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
611 }
Sage Weil124e68e2009-10-06 11:31:08 -0700612}
613
majianpenge8344e62013-09-12 13:54:26 +0800614
Sage Weil124e68e2009-10-06 11:31:08 -0700615/*
majianpenge8344e62013-09-12 13:54:26 +0800616 * Synchronous write, straight from __user pointer or user pages.
Sage Weil124e68e2009-10-06 11:31:08 -0700617 *
618 * If write spans object boundary, just do multiple writes. (For a
619 * correct atomic write, we should e.g. take write locks on all
620 * objects, rollback on failure, etc.)
621 */
majianpenge8344e62013-09-12 13:54:26 +0800622static ssize_t
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800623ceph_sync_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
624 struct ceph_snap_context *snapc)
Sage Weil124e68e2009-10-06 11:31:08 -0700625{
majianpenge8344e62013-09-12 13:54:26 +0800626 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -0500627 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700628 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700629 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Alex Elderacead002013-03-14 14:09:05 -0500630 struct ceph_vino vino;
Sage Weil124e68e2009-10-06 11:31:08 -0700631 struct ceph_osd_request *req;
632 struct page **pages;
633 int num_pages;
Sage Weil124e68e2009-10-06 11:31:08 -0700634 int written = 0;
635 int flags;
Sage Weil124e68e2009-10-06 11:31:08 -0700636 int check_caps = 0;
637 int ret;
638 struct timespec mtime = CURRENT_TIME;
Al Viro4908b822014-04-03 23:09:01 -0400639 size_t count = iov_iter_count(from);
Sage Weil124e68e2009-10-06 11:31:08 -0700640
Al Viro496ad9a2013-01-23 17:07:38 -0500641 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
Sage Weil124e68e2009-10-06 11:31:08 -0700642 return -EROFS;
643
majianpenge8344e62013-09-12 13:54:26 +0800644 dout("sync_direct_write on file %p %lld~%u\n", file, pos,
645 (unsigned)count);
Sage Weil124e68e2009-10-06 11:31:08 -0700646
majianpenge8344e62013-09-12 13:54:26 +0800647 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800648 if (ret < 0)
649 return ret;
650
651 ret = invalidate_inode_pages2_range(inode->i_mapping,
652 pos >> PAGE_CACHE_SHIFT,
majianpenge8344e62013-09-12 13:54:26 +0800653 (pos + count) >> PAGE_CACHE_SHIFT);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800654 if (ret < 0)
655 dout("invalidate_inode_pages2_range returned %d\n", ret);
656
Sage Weil124e68e2009-10-06 11:31:08 -0700657 flags = CEPH_OSD_FLAG_ORDERSNAP |
658 CEPH_OSD_FLAG_ONDISK |
659 CEPH_OSD_FLAG_WRITE;
Sage Weil124e68e2009-10-06 11:31:08 -0700660
Al Viro4908b822014-04-03 23:09:01 -0400661 while (iov_iter_count(from) > 0) {
Zhu, Caifengb5b98989d2015-10-08 15:26:15 +0800662 u64 len = dio_get_pagev_size(from);
Al Viro64c31312014-04-03 22:58:25 -0400663 size_t start;
664 ssize_t n;
majianpenge8344e62013-09-12 13:54:26 +0800665
majianpenge8344e62013-09-12 13:54:26 +0800666 vino = ceph_vino(inode);
667 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
Yan, Zheng715e4cd2014-11-13 14:40:37 +0800668 vino, pos, &len, 0,
majianpenge8344e62013-09-12 13:54:26 +0800669 2,/*include a 'startsync' command*/
670 CEPH_OSD_OP_WRITE, flags, snapc,
671 ci->i_truncate_seq,
672 ci->i_truncate_size,
673 false);
674 if (IS_ERR(req)) {
675 ret = PTR_ERR(req);
Al Viroeab87232014-04-03 22:44:19 -0400676 break;
majianpenge8344e62013-09-12 13:54:26 +0800677 }
678
Yan, Zheng144cba12015-04-27 11:09:54 +0800679 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0);
Yan, Zheng715e4cd2014-11-13 14:40:37 +0800680
Zhu, Caifengb5b98989d2015-10-08 15:26:15 +0800681 n = len;
682 pages = dio_get_pages_alloc(from, len, &start, &num_pages);
683 if (IS_ERR(pages)) {
Al Viro64c31312014-04-03 22:58:25 -0400684 ceph_osdc_put_request(req);
Zhu, Caifengb5b98989d2015-10-08 15:26:15 +0800685 ret = PTR_ERR(pages);
Al Viro64c31312014-04-03 22:58:25 -0400686 break;
Sage Weil124e68e2009-10-06 11:31:08 -0700687 }
688
689 /*
690 * throw out any page cache pages in this range. this
691 * may block.
692 */
Sage Weil213c99e2010-08-03 10:25:11 -0700693 truncate_inode_pages_range(inode->i_mapping, pos,
Al Viro64c31312014-04-03 22:58:25 -0400694 (pos+n) | (PAGE_CACHE_SIZE-1));
695 osd_req_op_extent_osd_data_pages(req, 0, pages, n, start,
majianpenge8344e62013-09-12 13:54:26 +0800696 false, false);
697
698 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
699 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
700
701 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
702 if (!ret)
703 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
704
705 ceph_put_page_vector(pages, num_pages, false);
706
majianpenge8344e62013-09-12 13:54:26 +0800707 ceph_osdc_put_request(req);
Al Viro64c31312014-04-03 22:58:25 -0400708 if (ret)
majianpenge8344e62013-09-12 13:54:26 +0800709 break;
Al Viro64c31312014-04-03 22:58:25 -0400710 pos += n;
711 written += n;
Al Viro4908b822014-04-03 23:09:01 -0400712 iov_iter_advance(from, n);
Al Viro64c31312014-04-03 22:58:25 -0400713
714 if (pos > i_size_read(inode)) {
715 check_caps = ceph_inode_set_size(inode, pos);
716 if (check_caps)
717 ceph_check_caps(ceph_inode(inode),
718 CHECK_CAPS_AUTHONLY,
719 NULL);
720 }
majianpenge8344e62013-09-12 13:54:26 +0800721 }
722
723 if (ret != -EOLDSNAPC && written > 0) {
724 iocb->ki_pos = pos;
725 ret = written;
726 }
727 return ret;
728}
729
730
731/*
732 * Synchronous write, straight from __user pointer or user pages.
733 *
734 * If write spans object boundary, just do multiple writes. (For a
735 * correct atomic write, we should e.g. take write locks on all
736 * objects, rollback on failure, etc.)
737 */
Yan, Zheng06fee302014-07-28 14:33:46 +0800738static ssize_t
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800739ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
740 struct ceph_snap_context *snapc)
majianpenge8344e62013-09-12 13:54:26 +0800741{
742 struct file *file = iocb->ki_filp;
743 struct inode *inode = file_inode(file);
744 struct ceph_inode_info *ci = ceph_inode(inode);
745 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
majianpenge8344e62013-09-12 13:54:26 +0800746 struct ceph_vino vino;
747 struct ceph_osd_request *req;
748 struct page **pages;
749 u64 len;
750 int num_pages;
751 int written = 0;
752 int flags;
753 int check_caps = 0;
754 int ret;
755 struct timespec mtime = CURRENT_TIME;
Al Viro4908b822014-04-03 23:09:01 -0400756 size_t count = iov_iter_count(from);
majianpenge8344e62013-09-12 13:54:26 +0800757
758 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
759 return -EROFS;
760
761 dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count);
762
763 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
764 if (ret < 0)
765 return ret;
766
767 ret = invalidate_inode_pages2_range(inode->i_mapping,
768 pos >> PAGE_CACHE_SHIFT,
769 (pos + count) >> PAGE_CACHE_SHIFT);
770 if (ret < 0)
771 dout("invalidate_inode_pages2_range returned %d\n", ret);
772
773 flags = CEPH_OSD_FLAG_ORDERSNAP |
774 CEPH_OSD_FLAG_ONDISK |
775 CEPH_OSD_FLAG_WRITE |
776 CEPH_OSD_FLAG_ACK;
777
Al Viro4908b822014-04-03 23:09:01 -0400778 while ((len = iov_iter_count(from)) > 0) {
majianpenge8344e62013-09-12 13:54:26 +0800779 size_t left;
780 int n;
781
majianpenge8344e62013-09-12 13:54:26 +0800782 vino = ceph_vino(inode);
783 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
Yan, Zheng715e4cd2014-11-13 14:40:37 +0800784 vino, pos, &len, 0, 1,
majianpenge8344e62013-09-12 13:54:26 +0800785 CEPH_OSD_OP_WRITE, flags, snapc,
786 ci->i_truncate_seq,
787 ci->i_truncate_size,
788 false);
789 if (IS_ERR(req)) {
790 ret = PTR_ERR(req);
Al Viroeab87232014-04-03 22:44:19 -0400791 break;
majianpenge8344e62013-09-12 13:54:26 +0800792 }
793
794 /*
795 * write from beginning of first page,
796 * regardless of io alignment
797 */
798 num_pages = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
799
Yan, Zheng687265e2015-06-13 17:27:05 +0800800 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
Sage Weil124e68e2009-10-06 11:31:08 -0700801 if (IS_ERR(pages)) {
802 ret = PTR_ERR(pages);
803 goto out;
804 }
majianpenge8344e62013-09-12 13:54:26 +0800805
806 left = len;
807 for (n = 0; n < num_pages; n++) {
Ilya Dryomov125d7252014-01-28 19:16:18 +0200808 size_t plen = min_t(size_t, left, PAGE_SIZE);
Al Viro4908b822014-04-03 23:09:01 -0400809 ret = copy_page_from_iter(pages[n], 0, plen, from);
majianpenge8344e62013-09-12 13:54:26 +0800810 if (ret != plen) {
811 ret = -EFAULT;
812 break;
813 }
814 left -= ret;
majianpenge8344e62013-09-12 13:54:26 +0800815 }
816
Sage Weil124e68e2009-10-06 11:31:08 -0700817 if (ret < 0) {
818 ceph_release_page_vector(pages, num_pages);
819 goto out;
820 }
821
majianpenge8344e62013-09-12 13:54:26 +0800822 /* get a second commit callback */
823 req->r_unsafe_callback = ceph_sync_write_unsafe;
824 req->r_inode = inode;
Sage Weil124e68e2009-10-06 11:31:08 -0700825
majianpenge8344e62013-09-12 13:54:26 +0800826 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
827 false, true);
Alex Elder02ee07d2013-03-14 14:09:06 -0500828
majianpenge8344e62013-09-12 13:54:26 +0800829 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
830 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
Sage Weil124e68e2009-10-06 11:31:08 -0700831
majianpenge8344e62013-09-12 13:54:26 +0800832 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
833 if (!ret)
834 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
Sage Weil124e68e2009-10-06 11:31:08 -0700835
836out:
majianpenge8344e62013-09-12 13:54:26 +0800837 ceph_osdc_put_request(req);
838 if (ret == 0) {
839 pos += len;
840 written += len;
Sage Weil124e68e2009-10-06 11:31:08 -0700841
majianpenge8344e62013-09-12 13:54:26 +0800842 if (pos > i_size_read(inode)) {
843 check_caps = ceph_inode_set_size(inode, pos);
844 if (check_caps)
845 ceph_check_caps(ceph_inode(inode),
846 CHECK_CAPS_AUTHONLY,
847 NULL);
848 }
849 } else
850 break;
851 }
852
853 if (ret != -EOLDSNAPC && written > 0) {
Sage Weil124e68e2009-10-06 11:31:08 -0700854 ret = written;
majianpenge8344e62013-09-12 13:54:26 +0800855 iocb->ki_pos = pos;
Sage Weil124e68e2009-10-06 11:31:08 -0700856 }
857 return ret;
858}
859
860/*
861 * Wrap generic_file_aio_read with checks for cap bits on the inode.
862 * Atomically grab references, so that those bits are not released
863 * back to the MDS mid-read.
864 *
865 * Hmm, the sync read case isn't actually async... should it be?
866 */
Al Viro36444242014-04-02 20:28:01 -0400867static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
Sage Weil124e68e2009-10-06 11:31:08 -0700868{
869 struct file *filp = iocb->ki_filp;
Sage Weil29625072010-05-27 10:40:43 -0700870 struct ceph_file_info *fi = filp->private_data;
Christoph Hellwig66ee59a2015-02-11 19:56:46 +0100871 size_t len = iov_iter_count(to);
Al Viro496ad9a2013-01-23 17:07:38 -0500872 struct inode *inode = file_inode(filp);
Sage Weil124e68e2009-10-06 11:31:08 -0700873 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng3738daa2014-11-14 22:10:07 +0800874 struct page *pinned_page = NULL;
Sage Weil124e68e2009-10-06 11:31:08 -0700875 ssize_t ret;
Sage Weil29625072010-05-27 10:40:43 -0700876 int want, got = 0;
Yan, Zheng83701242014-11-14 22:36:18 +0800877 int retry_op = 0, read = 0;
Sage Weil124e68e2009-10-06 11:31:08 -0700878
Sage Weil6a026582010-02-09 14:04:02 -0800879again:
majianpeng8eb4efb2013-09-26 14:42:17 +0800880 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
881 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
882
Sage Weil29625072010-05-27 10:40:43 -0700883 if (fi->fmode & CEPH_FILE_MODE_LAZY)
884 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
885 else
886 want = CEPH_CAP_FILE_CACHE;
Yan, Zheng3738daa2014-11-14 22:10:07 +0800887 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
Sage Weil124e68e2009-10-06 11:31:08 -0700888 if (ret < 0)
majianpeng8eb4efb2013-09-26 14:42:17 +0800889 return ret;
Sage Weil124e68e2009-10-06 11:31:08 -0700890
Sage Weil29625072010-05-27 10:40:43 -0700891 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Al Viro2ba48ce2015-04-09 13:52:01 -0400892 (iocb->ki_flags & IOCB_DIRECT) ||
majianpeng8eb4efb2013-09-26 14:42:17 +0800893 (fi->flags & CEPH_F_SYNC)) {
Sage Weil124e68e2009-10-06 11:31:08 -0700894
majianpeng8eb4efb2013-09-26 14:42:17 +0800895 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
896 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
897 ceph_cap_string(got));
898
Yan, Zheng83701242014-11-14 22:36:18 +0800899 if (ci->i_inline_version == CEPH_INLINE_NONE) {
900 /* hmm, this isn't really async... */
901 ret = ceph_sync_read(iocb, to, &retry_op);
902 } else {
903 retry_op = READ_INLINE;
904 }
majianpeng8eb4efb2013-09-26 14:42:17 +0800905 } else {
majianpeng8eb4efb2013-09-26 14:42:17 +0800906 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
Al Viro36444242014-04-02 20:28:01 -0400907 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
majianpeng8eb4efb2013-09-26 14:42:17 +0800908 ceph_cap_string(got));
909
Al Viro36444242014-04-02 20:28:01 -0400910 ret = generic_file_read_iter(iocb, to);
majianpeng8eb4efb2013-09-26 14:42:17 +0800911 }
Sage Weil124e68e2009-10-06 11:31:08 -0700912 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
913 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
Yan, Zheng3738daa2014-11-14 22:10:07 +0800914 if (pinned_page) {
915 page_cache_release(pinned_page);
916 pinned_page = NULL;
917 }
Sage Weil124e68e2009-10-06 11:31:08 -0700918 ceph_put_cap_refs(ci, got);
Yan, Zheng83701242014-11-14 22:36:18 +0800919 if (retry_op && ret >= 0) {
920 int statret;
921 struct page *page = NULL;
922 loff_t i_size;
923 if (retry_op == READ_INLINE) {
Yan, Zheng687265e2015-06-13 17:27:05 +0800924 page = __page_cache_alloc(GFP_KERNEL);
Yan, Zheng83701242014-11-14 22:36:18 +0800925 if (!page)
926 return -ENOMEM;
927 }
Sage Weil6a026582010-02-09 14:04:02 -0800928
Yan, Zheng83701242014-11-14 22:36:18 +0800929 statret = __ceph_do_getattr(inode, page,
930 CEPH_STAT_CAP_INLINE_DATA, !!page);
931 if (statret < 0) {
932 __free_page(page);
933 if (statret == -ENODATA) {
934 BUG_ON(retry_op != READ_INLINE);
935 goto again;
936 }
937 return statret;
938 }
939
940 i_size = i_size_read(inode);
941 if (retry_op == READ_INLINE) {
Yan, Zhengfcc02d22015-01-10 11:43:12 +0800942 BUG_ON(ret > 0 || read > 0);
943 if (iocb->ki_pos < i_size &&
944 iocb->ki_pos < PAGE_CACHE_SIZE) {
Yan, Zheng83701242014-11-14 22:36:18 +0800945 loff_t end = min_t(loff_t, i_size,
946 iocb->ki_pos + len);
Yan, Zhengfcc02d22015-01-10 11:43:12 +0800947 end = min_t(loff_t, end, PAGE_CACHE_SIZE);
Yan, Zheng83701242014-11-14 22:36:18 +0800948 if (statret < end)
949 zero_user_segment(page, statret, end);
950 ret = copy_page_to_iter(page,
951 iocb->ki_pos & ~PAGE_MASK,
952 end - iocb->ki_pos, to);
953 iocb->ki_pos += ret;
Yan, Zhengfcc02d22015-01-10 11:43:12 +0800954 read += ret;
955 }
956 if (iocb->ki_pos < i_size && read < len) {
957 size_t zlen = min_t(size_t, len - read,
958 i_size - iocb->ki_pos);
959 ret = iov_iter_zero(zlen, to);
960 iocb->ki_pos += ret;
961 read += ret;
Yan, Zheng83701242014-11-14 22:36:18 +0800962 }
963 __free_pages(page, 0);
Yan, Zhengfcc02d22015-01-10 11:43:12 +0800964 return read;
Yan, Zheng83701242014-11-14 22:36:18 +0800965 }
Sage Weil6a026582010-02-09 14:04:02 -0800966
967 /* hit EOF or hole? */
Yan, Zheng83701242014-11-14 22:36:18 +0800968 if (retry_op == CHECK_EOF && iocb->ki_pos < i_size &&
Yan, Zhengfcc02d22015-01-10 11:43:12 +0800969 ret < len) {
majianpeng8eb4efb2013-09-26 14:42:17 +0800970 dout("sync_read hit hole, ppos %lld < size %lld"
971 ", reading more\n", iocb->ki_pos,
972 inode->i_size);
973
Sage Weil6a026582010-02-09 14:04:02 -0800974 read += ret;
Sage Weil6a026582010-02-09 14:04:02 -0800975 len -= ret;
Yan, Zheng83701242014-11-14 22:36:18 +0800976 retry_op = 0;
Sage Weil6a026582010-02-09 14:04:02 -0800977 goto again;
978 }
979 }
majianpeng8eb4efb2013-09-26 14:42:17 +0800980
Sage Weil6a026582010-02-09 14:04:02 -0800981 if (ret >= 0)
982 ret += read;
983
Sage Weil124e68e2009-10-06 11:31:08 -0700984 return ret;
985}
986
987/*
988 * Take cap references to avoid releasing caps to MDS mid-write.
989 *
990 * If we are synchronous, and write with an old snap context, the OSD
991 * may return EOLDSNAPC. In that case, retry the write.. _after_
992 * dropping our cap refs and allowing the pending snap to logically
993 * complete _before_ this write occurs.
994 *
995 * If we are near ENOSPC, write synchronously.
996 */
Al Viro4908b822014-04-03 23:09:01 -0400997static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
Sage Weil124e68e2009-10-06 11:31:08 -0700998{
999 struct file *file = iocb->ki_filp;
Sage Weil33caad32010-05-26 14:31:27 -07001000 struct ceph_file_info *fi = file->private_data;
Al Viro496ad9a2013-01-23 17:07:38 -05001001 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -07001002 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001003 struct ceph_osd_client *osdc =
1004 &ceph_sb_to_client(inode->i_sb)->client->osdc;
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001005 struct ceph_cap_flush *prealloc_cf;
Al Viro3309dd02015-04-09 12:55:47 -04001006 ssize_t count, written = 0;
Yan, Zheng03d254e2013-04-12 16:11:13 +08001007 int err, want, got;
Al Viro3309dd02015-04-09 12:55:47 -04001008 loff_t pos;
Sage Weil124e68e2009-10-06 11:31:08 -07001009
1010 if (ceph_snap(inode) != CEPH_NOSNAP)
1011 return -EROFS;
1012
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001013 prealloc_cf = ceph_alloc_cap_flush();
1014 if (!prealloc_cf)
1015 return -ENOMEM;
1016
Yan, Zheng03d254e2013-04-12 16:11:13 +08001017 mutex_lock(&inode->i_mutex);
Yan, Zheng03d254e2013-04-12 16:11:13 +08001018
Yan, Zheng03d254e2013-04-12 16:11:13 +08001019 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01001020 current->backing_dev_info = inode_to_bdi(inode);
Yan, Zheng03d254e2013-04-12 16:11:13 +08001021
Yan, Zheng55b0b312015-09-07 11:35:01 +08001022 if (iocb->ki_flags & IOCB_APPEND) {
1023 err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1024 if (err < 0)
1025 goto out;
1026 }
1027
Al Viro3309dd02015-04-09 12:55:47 -04001028 err = generic_write_checks(iocb, from);
1029 if (err <= 0)
Yan, Zheng03d254e2013-04-12 16:11:13 +08001030 goto out;
1031
Al Viro3309dd02015-04-09 12:55:47 -04001032 pos = iocb->ki_pos;
1033 count = iov_iter_count(from);
Jan Kara5fa8e0a2015-05-21 16:05:53 +02001034 err = file_remove_privs(file);
Yan, Zheng03d254e2013-04-12 16:11:13 +08001035 if (err)
1036 goto out;
1037
1038 err = file_update_time(file);
1039 if (err)
1040 goto out;
1041
Yan, Zheng28127bd2014-11-14 22:38:29 +08001042 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1043 err = ceph_uninline_data(file, NULL);
1044 if (err < 0)
1045 goto out;
1046 }
1047
Sage Weil124e68e2009-10-06 11:31:08 -07001048retry_snap:
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001049 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) {
Yan, Zheng03d254e2013-04-12 16:11:13 +08001050 err = -ENOSPC;
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001051 goto out;
1052 }
Yan, Zheng03d254e2013-04-12 16:11:13 +08001053
Randy Dunlapac7f29b2013-04-19 14:20:07 -07001054 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
Yan, Zheng03d254e2013-04-12 16:11:13 +08001055 inode, ceph_vinop(inode), pos, count, inode->i_size);
Sage Weil7971bd92013-05-01 21:15:58 -07001056 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1057 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1058 else
1059 want = CEPH_CAP_FILE_BUFFER;
Yan, Zheng03d254e2013-04-12 16:11:13 +08001060 got = 0;
Yan, Zheng3738daa2014-11-14 22:10:07 +08001061 err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count,
1062 &got, NULL);
Yan, Zheng03d254e2013-04-12 16:11:13 +08001063 if (err < 0)
Yan, Zheng37505d52013-04-12 16:11:10 +08001064 goto out;
Sage Weil124e68e2009-10-06 11:31:08 -07001065
Randy Dunlapac7f29b2013-04-19 14:20:07 -07001066 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
Yan, Zheng03d254e2013-04-12 16:11:13 +08001067 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
Sage Weil7971bd92013-05-01 21:15:58 -07001068
1069 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Al Viro2ba48ce2015-04-09 13:52:01 -04001070 (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC)) {
Yan, Zheng5dda377c2015-04-30 14:40:54 +08001071 struct ceph_snap_context *snapc;
Al Viro4908b822014-04-03 23:09:01 -04001072 struct iov_iter data;
Yan, Zheng37505d52013-04-12 16:11:10 +08001073 mutex_unlock(&inode->i_mutex);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08001074
1075 spin_lock(&ci->i_ceph_lock);
1076 if (__ceph_have_pending_cap_snap(ci)) {
1077 struct ceph_cap_snap *capsnap =
1078 list_last_entry(&ci->i_cap_snaps,
1079 struct ceph_cap_snap,
1080 ci_item);
1081 snapc = ceph_get_snap_context(capsnap->context);
1082 } else {
1083 BUG_ON(!ci->i_head_snapc);
1084 snapc = ceph_get_snap_context(ci->i_head_snapc);
1085 }
1086 spin_unlock(&ci->i_ceph_lock);
1087
Al Viro4908b822014-04-03 23:09:01 -04001088 /* we might need to revert back to that point */
1089 data = *from;
Al Viro2ba48ce2015-04-09 13:52:01 -04001090 if (iocb->ki_flags & IOCB_DIRECT)
Yan, Zheng5dda377c2015-04-30 14:40:54 +08001091 written = ceph_sync_direct_write(iocb, &data, pos,
1092 snapc);
majianpenge8344e62013-09-12 13:54:26 +08001093 else
Yan, Zheng5dda377c2015-04-30 14:40:54 +08001094 written = ceph_sync_write(iocb, &data, pos, snapc);
majianpeng0e5dd452013-08-08 15:32:19 +08001095 if (written == -EOLDSNAPC) {
1096 dout("aio_write %p %llx.%llx %llu~%u"
1097 "got EOLDSNAPC, retrying\n",
1098 inode, ceph_vinop(inode),
Al Viro4908b822014-04-03 23:09:01 -04001099 pos, (unsigned)count);
majianpeng0e5dd452013-08-08 15:32:19 +08001100 mutex_lock(&inode->i_mutex);
majianpeng0e5dd452013-08-08 15:32:19 +08001101 goto retry_snap;
1102 }
Al Viro4908b822014-04-03 23:09:01 -04001103 if (written > 0)
1104 iov_iter_advance(from, written);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08001105 ceph_put_snap_context(snapc);
Sage Weil7971bd92013-05-01 21:15:58 -07001106 } else {
Yunchuan Wen32d3e1482013-12-26 06:29:27 -08001107 loff_t old_size = inode->i_size;
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001108 /*
1109 * No need to acquire the i_truncate_mutex. Because
1110 * the MDS revokes Fwb caps before sending truncate
1111 * message to us. We can't get Fwb cap while there
1112 * are pending vmtruncate. So write and vmtruncate
1113 * can not run at the same time
1114 */
Al Viro4908b822014-04-03 23:09:01 -04001115 written = generic_perform_write(file, from, pos);
Al Viroaec605f42014-02-11 22:28:43 -05001116 if (likely(written >= 0))
1117 iocb->ki_pos = pos + written;
Yunchuan Wen32d3e1482013-12-26 06:29:27 -08001118 if (inode->i_size > old_size)
1119 ceph_fscache_update_objectsize(inode);
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001120 mutex_unlock(&inode->i_mutex);
Sage Weil124e68e2009-10-06 11:31:08 -07001121 }
Sage Weild8de9ab2011-07-26 11:27:34 -07001122
Yan, Zheng03d254e2013-04-12 16:11:13 +08001123 if (written >= 0) {
Sage Weilfca65b42011-05-04 11:33:47 -07001124 int dirty;
Sage Weilbe655592011-11-30 09:47:09 -08001125 spin_lock(&ci->i_ceph_lock);
Yan, Zheng28127bd2014-11-14 22:38:29 +08001126 ci->i_inline_version = CEPH_INLINE_NONE;
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001127 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1128 &prealloc_cf);
Sage Weilbe655592011-11-30 09:47:09 -08001129 spin_unlock(&ci->i_ceph_lock);
Sage Weilfca65b42011-05-04 11:33:47 -07001130 if (dirty)
1131 __mark_inode_dirty(inode, dirty);
Sage Weil124e68e2009-10-06 11:31:08 -07001132 }
Sage Weil7971bd92013-05-01 21:15:58 -07001133
Sage Weil124e68e2009-10-06 11:31:08 -07001134 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
Al Viro4908b822014-04-03 23:09:01 -04001135 inode, ceph_vinop(inode), pos, (unsigned)count,
Sage Weil7971bd92013-05-01 21:15:58 -07001136 ceph_cap_string(got));
Sage Weil124e68e2009-10-06 11:31:08 -07001137 ceph_put_cap_refs(ci, got);
Sage Weil7971bd92013-05-01 21:15:58 -07001138
Yan, Zheng03d254e2013-04-12 16:11:13 +08001139 if (written >= 0 &&
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001140 ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) ||
1141 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) {
Yan, Zheng03d254e2013-04-12 16:11:13 +08001142 err = vfs_fsync_range(file, pos, pos + written - 1, 1);
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001143 if (err < 0)
Yan, Zheng03d254e2013-04-12 16:11:13 +08001144 written = err;
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001145 }
Yan, Zheng03d254e2013-04-12 16:11:13 +08001146
Sage Weil2f75e9e2013-08-09 09:57:58 -07001147 goto out_unlocked;
Sage Weil124e68e2009-10-06 11:31:08 -07001148
Sage Weil2f75e9e2013-08-09 09:57:58 -07001149out:
1150 mutex_unlock(&inode->i_mutex);
1151out_unlocked:
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001152 ceph_free_cap_flush(prealloc_cf);
Sage Weil2f75e9e2013-08-09 09:57:58 -07001153 current->backing_dev_info = NULL;
Yan, Zheng03d254e2013-04-12 16:11:13 +08001154 return written ? written : err;
Sage Weil124e68e2009-10-06 11:31:08 -07001155}
1156
1157/*
1158 * llseek. be sure to verify file size on SEEK_END.
1159 */
Andrew Morton965c8e52012-12-17 15:59:39 -08001160static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
Sage Weil124e68e2009-10-06 11:31:08 -07001161{
1162 struct inode *inode = file->f_mapping->host;
1163 int ret;
1164
1165 mutex_lock(&inode->i_mutex);
Sage Weil6a82c472011-12-13 09:19:26 -08001166
Andrew Morton965c8e52012-12-17 15:59:39 -08001167 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
Yan, Zheng508b32d2014-09-16 21:46:17 +08001168 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
Sage Weil124e68e2009-10-06 11:31:08 -07001169 if (ret < 0) {
1170 offset = ret;
1171 goto out;
1172 }
Josef Bacik06222e42011-07-18 13:21:38 -04001173 }
1174
Andrew Morton965c8e52012-12-17 15:59:39 -08001175 switch (whence) {
Josef Bacik06222e42011-07-18 13:21:38 -04001176 case SEEK_END:
Sage Weil124e68e2009-10-06 11:31:08 -07001177 offset += inode->i_size;
1178 break;
1179 case SEEK_CUR:
1180 /*
1181 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1182 * position-querying operation. Avoid rewriting the "same"
1183 * f_pos value back to the file because a concurrent read(),
1184 * write() or lseek() might have altered it
1185 */
1186 if (offset == 0) {
1187 offset = file->f_pos;
1188 goto out;
1189 }
1190 offset += file->f_pos;
1191 break;
Josef Bacik06222e42011-07-18 13:21:38 -04001192 case SEEK_DATA:
1193 if (offset >= inode->i_size) {
1194 ret = -ENXIO;
1195 goto out;
1196 }
1197 break;
1198 case SEEK_HOLE:
1199 if (offset >= inode->i_size) {
1200 ret = -ENXIO;
1201 goto out;
1202 }
1203 offset = inode->i_size;
1204 break;
Sage Weil124e68e2009-10-06 11:31:08 -07001205 }
1206
Jie Liu46a1c2c72013-06-25 12:02:13 +08001207 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Sage Weil124e68e2009-10-06 11:31:08 -07001208
1209out:
1210 mutex_unlock(&inode->i_mutex);
1211 return offset;
1212}
1213
Li Wangad7a60de2013-08-15 11:51:44 +08001214static inline void ceph_zero_partial_page(
1215 struct inode *inode, loff_t offset, unsigned size)
1216{
1217 struct page *page;
1218 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
1219
1220 page = find_lock_page(inode->i_mapping, index);
1221 if (page) {
1222 wait_on_page_writeback(page);
1223 zero_user(page, offset & (PAGE_CACHE_SIZE - 1), size);
1224 unlock_page(page);
1225 page_cache_release(page);
1226 }
1227}
1228
1229static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1230 loff_t length)
1231{
1232 loff_t nearly = round_up(offset, PAGE_CACHE_SIZE);
1233 if (offset < nearly) {
1234 loff_t size = nearly - offset;
1235 if (length < size)
1236 size = length;
1237 ceph_zero_partial_page(inode, offset, size);
1238 offset += size;
1239 length -= size;
1240 }
1241 if (length >= PAGE_CACHE_SIZE) {
1242 loff_t size = round_down(length, PAGE_CACHE_SIZE);
1243 truncate_pagecache_range(inode, offset, offset + size - 1);
1244 offset += size;
1245 length -= size;
1246 }
1247 if (length)
1248 ceph_zero_partial_page(inode, offset, length);
1249}
1250
1251static int ceph_zero_partial_object(struct inode *inode,
1252 loff_t offset, loff_t *length)
1253{
1254 struct ceph_inode_info *ci = ceph_inode(inode);
1255 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1256 struct ceph_osd_request *req;
1257 int ret = 0;
1258 loff_t zero = 0;
1259 int op;
1260
1261 if (!length) {
1262 op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1263 length = &zero;
1264 } else {
1265 op = CEPH_OSD_OP_ZERO;
1266 }
1267
1268 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1269 ceph_vino(inode),
1270 offset, length,
Yan, Zheng715e4cd2014-11-13 14:40:37 +08001271 0, 1, op,
Li Wangad7a60de2013-08-15 11:51:44 +08001272 CEPH_OSD_FLAG_WRITE |
1273 CEPH_OSD_FLAG_ONDISK,
1274 NULL, 0, 0, false);
1275 if (IS_ERR(req)) {
1276 ret = PTR_ERR(req);
1277 goto out;
1278 }
1279
1280 ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap,
1281 &inode->i_mtime);
1282
1283 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1284 if (!ret) {
1285 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1286 if (ret == -ENOENT)
1287 ret = 0;
1288 }
1289 ceph_osdc_put_request(req);
1290
1291out:
1292 return ret;
1293}
1294
1295static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1296{
1297 int ret = 0;
1298 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weilb314a902013-08-27 12:15:16 -07001299 s32 stripe_unit = ceph_file_layout_su(ci->i_layout);
1300 s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout);
1301 s32 object_size = ceph_file_layout_object_size(ci->i_layout);
1302 u64 object_set_size = object_size * stripe_count;
1303 u64 nearly, t;
Li Wangad7a60de2013-08-15 11:51:44 +08001304
Sage Weilb314a902013-08-27 12:15:16 -07001305 /* round offset up to next period boundary */
1306 nearly = offset + object_set_size - 1;
1307 t = nearly;
1308 nearly -= do_div(t, object_set_size);
1309
Li Wangad7a60de2013-08-15 11:51:44 +08001310 while (length && offset < nearly) {
1311 loff_t size = length;
1312 ret = ceph_zero_partial_object(inode, offset, &size);
1313 if (ret < 0)
1314 return ret;
1315 offset += size;
1316 length -= size;
1317 }
1318 while (length >= object_set_size) {
1319 int i;
1320 loff_t pos = offset;
1321 for (i = 0; i < stripe_count; ++i) {
1322 ret = ceph_zero_partial_object(inode, pos, NULL);
1323 if (ret < 0)
1324 return ret;
1325 pos += stripe_unit;
1326 }
1327 offset += object_set_size;
1328 length -= object_set_size;
1329 }
1330 while (length) {
1331 loff_t size = length;
1332 ret = ceph_zero_partial_object(inode, offset, &size);
1333 if (ret < 0)
1334 return ret;
1335 offset += size;
1336 length -= size;
1337 }
1338 return ret;
1339}
1340
1341static long ceph_fallocate(struct file *file, int mode,
1342 loff_t offset, loff_t length)
1343{
1344 struct ceph_file_info *fi = file->private_data;
Libo Chenaa8b60e2013-12-11 13:49:11 +08001345 struct inode *inode = file_inode(file);
Li Wangad7a60de2013-08-15 11:51:44 +08001346 struct ceph_inode_info *ci = ceph_inode(inode);
1347 struct ceph_osd_client *osdc =
1348 &ceph_inode_to_client(inode)->client->osdc;
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001349 struct ceph_cap_flush *prealloc_cf;
Li Wangad7a60de2013-08-15 11:51:44 +08001350 int want, got = 0;
1351 int dirty;
1352 int ret = 0;
1353 loff_t endoff = 0;
1354 loff_t size;
1355
Yan, Zheng494d77b2014-06-26 15:25:17 +08001356 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1357 return -EOPNOTSUPP;
1358
Li Wangad7a60de2013-08-15 11:51:44 +08001359 if (!S_ISREG(inode->i_mode))
1360 return -EOPNOTSUPP;
1361
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001362 prealloc_cf = ceph_alloc_cap_flush();
1363 if (!prealloc_cf)
1364 return -ENOMEM;
1365
Li Wangad7a60de2013-08-15 11:51:44 +08001366 mutex_lock(&inode->i_mutex);
1367
1368 if (ceph_snap(inode) != CEPH_NOSNAP) {
1369 ret = -EROFS;
1370 goto unlock;
1371 }
1372
1373 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) &&
1374 !(mode & FALLOC_FL_PUNCH_HOLE)) {
1375 ret = -ENOSPC;
1376 goto unlock;
1377 }
1378
Yan, Zheng28127bd2014-11-14 22:38:29 +08001379 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1380 ret = ceph_uninline_data(file, NULL);
1381 if (ret < 0)
1382 goto unlock;
1383 }
1384
Li Wangad7a60de2013-08-15 11:51:44 +08001385 size = i_size_read(inode);
1386 if (!(mode & FALLOC_FL_KEEP_SIZE))
1387 endoff = offset + length;
1388
1389 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1390 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1391 else
1392 want = CEPH_CAP_FILE_BUFFER;
1393
Yan, Zheng3738daa2014-11-14 22:10:07 +08001394 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL);
Li Wangad7a60de2013-08-15 11:51:44 +08001395 if (ret < 0)
1396 goto unlock;
1397
1398 if (mode & FALLOC_FL_PUNCH_HOLE) {
1399 if (offset < size)
1400 ceph_zero_pagecache_range(inode, offset, length);
1401 ret = ceph_zero_objects(inode, offset, length);
1402 } else if (endoff > size) {
1403 truncate_pagecache_range(inode, size, -1);
1404 if (ceph_inode_set_size(inode, endoff))
1405 ceph_check_caps(ceph_inode(inode),
1406 CHECK_CAPS_AUTHONLY, NULL);
1407 }
1408
1409 if (!ret) {
1410 spin_lock(&ci->i_ceph_lock);
Yan, Zheng28127bd2014-11-14 22:38:29 +08001411 ci->i_inline_version = CEPH_INLINE_NONE;
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001412 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1413 &prealloc_cf);
Li Wangad7a60de2013-08-15 11:51:44 +08001414 spin_unlock(&ci->i_ceph_lock);
1415 if (dirty)
1416 __mark_inode_dirty(inode, dirty);
1417 }
1418
1419 ceph_put_cap_refs(ci, got);
1420unlock:
1421 mutex_unlock(&inode->i_mutex);
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001422 ceph_free_cap_flush(prealloc_cf);
Li Wangad7a60de2013-08-15 11:51:44 +08001423 return ret;
1424}
1425
Sage Weil124e68e2009-10-06 11:31:08 -07001426const struct file_operations ceph_file_fops = {
1427 .open = ceph_open,
1428 .release = ceph_release,
1429 .llseek = ceph_llseek,
Al Viro36444242014-04-02 20:28:01 -04001430 .read_iter = ceph_read_iter,
Al Viro4908b822014-04-03 23:09:01 -04001431 .write_iter = ceph_write_iter,
Sage Weil124e68e2009-10-06 11:31:08 -07001432 .mmap = ceph_mmap,
1433 .fsync = ceph_fsync,
Greg Farnum40819f62010-08-02 15:34:23 -07001434 .lock = ceph_lock,
1435 .flock = ceph_flock,
Sage Weil124e68e2009-10-06 11:31:08 -07001436 .splice_read = generic_file_splice_read,
Al Viro3551dd72014-04-05 04:40:12 -04001437 .splice_write = iter_file_splice_write,
Sage Weil124e68e2009-10-06 11:31:08 -07001438 .unlocked_ioctl = ceph_ioctl,
1439 .compat_ioctl = ceph_ioctl,
Li Wangad7a60de2013-08-15 11:51:44 +08001440 .fallocate = ceph_fallocate,
Sage Weil124e68e2009-10-06 11:31:08 -07001441};
1442