Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/fs/nfs/read.c |
| 3 | * |
| 4 | * Block I/O for NFS |
| 5 | * |
| 6 | * Partial copy of Linus' read cache modifications to fs/nfs/file.c |
| 7 | * modified for async RPC by okir@monad.swb.de |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 8 | */ |
| 9 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 10 | #include <linux/time.h> |
| 11 | #include <linux/kernel.h> |
| 12 | #include <linux/errno.h> |
| 13 | #include <linux/fcntl.h> |
| 14 | #include <linux/stat.h> |
| 15 | #include <linux/mm.h> |
| 16 | #include <linux/slab.h> |
| 17 | #include <linux/pagemap.h> |
| 18 | #include <linux/sunrpc/clnt.h> |
| 19 | #include <linux/nfs_fs.h> |
| 20 | #include <linux/nfs_page.h> |
| 21 | #include <linux/smp_lock.h> |
| 22 | |
| 23 | #include <asm/system.h> |
| 24 | |
Trond Myklebust | 49a70f2 | 2006-12-05 00:35:38 -0500 | [diff] [blame] | 25 | #include "internal.h" |
Chuck Lever | 91d5b47 | 2006-03-20 13:44:14 -0500 | [diff] [blame] | 26 | #include "iostat.h" |
| 27 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
| 29 | |
Trond Myklebust | 8d5658c | 2007-04-10 09:26:35 -0400 | [diff] [blame] | 30 | static int nfs_pagein_multi(struct inode *, struct list_head *, unsigned int, size_t, int); |
| 31 | static int nfs_pagein_one(struct inode *, struct list_head *, unsigned int, size_t, int); |
Trond Myklebust | ec06c09 | 2006-03-20 13:44:27 -0500 | [diff] [blame] | 32 | static const struct rpc_call_ops nfs_read_partial_ops; |
| 33 | static const struct rpc_call_ops nfs_read_full_ops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 35 | static struct kmem_cache *nfs_rdata_cachep; |
Trond Myklebust | 3feb2d4 | 2006-03-20 13:44:37 -0500 | [diff] [blame] | 36 | static mempool_t *nfs_rdata_mempool; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 37 | |
| 38 | #define MIN_POOL_READ (32) |
| 39 | |
Trond Myklebust | 8d5658c | 2007-04-10 09:26:35 -0400 | [diff] [blame] | 40 | struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount) |
Trond Myklebust | 3feb2d4 | 2006-03-20 13:44:37 -0500 | [diff] [blame] | 41 | { |
Christoph Lameter | e6b4f8d | 2006-12-06 20:33:14 -0800 | [diff] [blame] | 42 | struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_NOFS); |
Trond Myklebust | 3feb2d4 | 2006-03-20 13:44:37 -0500 | [diff] [blame] | 43 | |
| 44 | if (p) { |
| 45 | memset(p, 0, sizeof(*p)); |
| 46 | INIT_LIST_HEAD(&p->pages); |
Trond Myklebust | e9f7bee | 2006-09-08 09:48:54 -0700 | [diff] [blame] | 47 | p->npages = pagecount; |
Chuck Lever | 0d0b5cb | 2006-05-25 01:40:53 -0400 | [diff] [blame] | 48 | if (pagecount <= ARRAY_SIZE(p->page_array)) |
| 49 | p->pagevec = p->page_array; |
Trond Myklebust | 3feb2d4 | 2006-03-20 13:44:37 -0500 | [diff] [blame] | 50 | else { |
Chuck Lever | 0d0b5cb | 2006-05-25 01:40:53 -0400 | [diff] [blame] | 51 | p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); |
| 52 | if (!p->pagevec) { |
Trond Myklebust | 3feb2d4 | 2006-03-20 13:44:37 -0500 | [diff] [blame] | 53 | mempool_free(p, nfs_rdata_mempool); |
| 54 | p = NULL; |
| 55 | } |
| 56 | } |
| 57 | } |
| 58 | return p; |
| 59 | } |
| 60 | |
Trond Myklebust | 8aca67f | 2006-11-13 16:23:44 -0500 | [diff] [blame] | 61 | static void nfs_readdata_rcu_free(struct rcu_head *head) |
Trond Myklebust | 3feb2d4 | 2006-03-20 13:44:37 -0500 | [diff] [blame] | 62 | { |
Trond Myklebust | 8aca67f | 2006-11-13 16:23:44 -0500 | [diff] [blame] | 63 | struct nfs_read_data *p = container_of(head, struct nfs_read_data, task.u.tk_rcu); |
Trond Myklebust | 3feb2d4 | 2006-03-20 13:44:37 -0500 | [diff] [blame] | 64 | if (p && (p->pagevec != &p->page_array[0])) |
| 65 | kfree(p->pagevec); |
| 66 | mempool_free(p, nfs_rdata_mempool); |
| 67 | } |
| 68 | |
Trond Myklebust | 8aca67f | 2006-11-13 16:23:44 -0500 | [diff] [blame] | 69 | static void nfs_readdata_free(struct nfs_read_data *rdata) |
| 70 | { |
| 71 | call_rcu_bh(&rdata->task.u.tk_rcu, nfs_readdata_rcu_free); |
| 72 | } |
| 73 | |
Trond Myklebust | 963d8fe | 2006-01-03 09:55:04 +0100 | [diff] [blame] | 74 | void nfs_readdata_release(void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 75 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 76 | nfs_readdata_free(data); |
| 77 | } |
| 78 | |
| 79 | static |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 80 | int nfs_return_empty_page(struct page *page) |
| 81 | { |
Christoph Lameter | eebd2aa | 2008-02-04 22:28:29 -0800 | [diff] [blame] | 82 | zero_user(page, 0, PAGE_CACHE_SIZE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 83 | SetPageUptodate(page); |
| 84 | unlock_page(page); |
| 85 | return 0; |
| 86 | } |
| 87 | |
Trond Myklebust | 1de3fc1 | 2006-05-25 01:40:44 -0400 | [diff] [blame] | 88 | static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data) |
| 89 | { |
| 90 | unsigned int remainder = data->args.count - data->res.count; |
| 91 | unsigned int base = data->args.pgbase + data->res.count; |
| 92 | unsigned int pglen; |
| 93 | struct page **pages; |
| 94 | |
| 95 | if (data->res.eof == 0 || remainder == 0) |
| 96 | return; |
| 97 | /* |
| 98 | * Note: "remainder" can never be negative, since we check for |
| 99 | * this in the XDR code. |
| 100 | */ |
| 101 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; |
| 102 | base &= ~PAGE_CACHE_MASK; |
| 103 | pglen = PAGE_CACHE_SIZE - base; |
Trond Myklebust | 79558f3 | 2006-08-22 13:44:32 -0400 | [diff] [blame] | 104 | for (;;) { |
| 105 | if (remainder <= pglen) { |
Christoph Lameter | eebd2aa | 2008-02-04 22:28:29 -0800 | [diff] [blame] | 106 | zero_user(*pages, base, remainder); |
Trond Myklebust | 79558f3 | 2006-08-22 13:44:32 -0400 | [diff] [blame] | 107 | break; |
| 108 | } |
Christoph Lameter | eebd2aa | 2008-02-04 22:28:29 -0800 | [diff] [blame] | 109 | zero_user(*pages, base, pglen); |
Trond Myklebust | 79558f3 | 2006-08-22 13:44:32 -0400 | [diff] [blame] | 110 | pages++; |
| 111 | remainder -= pglen; |
| 112 | pglen = PAGE_CACHE_SIZE; |
| 113 | base = 0; |
| 114 | } |
Trond Myklebust | 1de3fc1 | 2006-05-25 01:40:44 -0400 | [diff] [blame] | 115 | } |
| 116 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 117 | static int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode, |
| 118 | struct page *page) |
| 119 | { |
| 120 | LIST_HEAD(one_request); |
| 121 | struct nfs_page *new; |
| 122 | unsigned int len; |
| 123 | |
Trond Myklebust | 49a70f2 | 2006-12-05 00:35:38 -0500 | [diff] [blame] | 124 | len = nfs_page_length(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 125 | if (len == 0) |
| 126 | return nfs_return_empty_page(page); |
| 127 | new = nfs_create_request(ctx, inode, page, 0, len); |
| 128 | if (IS_ERR(new)) { |
| 129 | unlock_page(page); |
| 130 | return PTR_ERR(new); |
| 131 | } |
| 132 | if (len < PAGE_CACHE_SIZE) |
Christoph Lameter | eebd2aa | 2008-02-04 22:28:29 -0800 | [diff] [blame] | 133 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 134 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 135 | nfs_list_add_request(new, &one_request); |
Trond Myklebust | bcb71bb | 2007-04-02 18:48:28 -0400 | [diff] [blame] | 136 | if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE) |
Trond Myklebust | 8d5658c | 2007-04-10 09:26:35 -0400 | [diff] [blame] | 137 | nfs_pagein_multi(inode, &one_request, 1, len, 0); |
Trond Myklebust | bcb71bb | 2007-04-02 18:48:28 -0400 | [diff] [blame] | 138 | else |
Trond Myklebust | 8d5658c | 2007-04-10 09:26:35 -0400 | [diff] [blame] | 139 | nfs_pagein_one(inode, &one_request, 1, len, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 140 | return 0; |
| 141 | } |
| 142 | |
| 143 | static void nfs_readpage_release(struct nfs_page *req) |
| 144 | { |
| 145 | unlock_page(req->wb_page); |
| 146 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 147 | dprintk("NFS: read done (%s/%Ld %d@%Ld)\n", |
Trond Myklebust | 88be9f9 | 2007-06-05 10:42:27 -0400 | [diff] [blame] | 148 | req->wb_context->path.dentry->d_inode->i_sb->s_id, |
| 149 | (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 150 | req->wb_bytes, |
| 151 | (long long)req_offset(req)); |
Nick Wilson | 10d2c46 | 2005-09-22 21:44:28 -0700 | [diff] [blame] | 152 | nfs_clear_request(req); |
| 153 | nfs_release_request(req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 154 | } |
| 155 | |
| 156 | /* |
| 157 | * Set up the NFS read request struct |
| 158 | */ |
| 159 | static void nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data, |
Trond Myklebust | ec06c09 | 2006-03-20 13:44:27 -0500 | [diff] [blame] | 160 | const struct rpc_call_ops *call_ops, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | unsigned int count, unsigned int offset) |
| 162 | { |
Trond Myklebust | 84115e1 | 2007-07-14 15:39:59 -0400 | [diff] [blame] | 163 | struct inode *inode = req->wb_context->path.dentry->d_inode; |
| 164 | int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0; |
Trond Myklebust | 0773769 | 2007-10-25 18:42:54 -0400 | [diff] [blame] | 165 | struct rpc_task *task; |
Trond Myklebust | bdc7f02 | 2007-07-14 15:40:00 -0400 | [diff] [blame] | 166 | struct rpc_message msg = { |
| 167 | .rpc_argp = &data->args, |
| 168 | .rpc_resp = &data->res, |
| 169 | .rpc_cred = req->wb_context->cred, |
| 170 | }; |
Trond Myklebust | 84115e1 | 2007-07-14 15:39:59 -0400 | [diff] [blame] | 171 | struct rpc_task_setup task_setup_data = { |
Trond Myklebust | 0773769 | 2007-10-25 18:42:54 -0400 | [diff] [blame] | 172 | .task = &data->task, |
Trond Myklebust | 84115e1 | 2007-07-14 15:39:59 -0400 | [diff] [blame] | 173 | .rpc_client = NFS_CLIENT(inode), |
Trond Myklebust | bdc7f02 | 2007-07-14 15:40:00 -0400 | [diff] [blame] | 174 | .rpc_message = &msg, |
Trond Myklebust | 84115e1 | 2007-07-14 15:39:59 -0400 | [diff] [blame] | 175 | .callback_ops = call_ops, |
| 176 | .callback_data = data, |
| 177 | .flags = RPC_TASK_ASYNC | swap_flags, |
| 178 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 179 | |
| 180 | data->req = req; |
Trond Myklebust | 84115e1 | 2007-07-14 15:39:59 -0400 | [diff] [blame] | 181 | data->inode = inode; |
Trond Myklebust | bdc7f02 | 2007-07-14 15:40:00 -0400 | [diff] [blame] | 182 | data->cred = msg.rpc_cred; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 183 | |
| 184 | data->args.fh = NFS_FH(inode); |
| 185 | data->args.offset = req_offset(req) + offset; |
| 186 | data->args.pgbase = req->wb_pgbase + offset; |
| 187 | data->args.pages = data->pagevec; |
| 188 | data->args.count = count; |
| 189 | data->args.context = req->wb_context; |
| 190 | |
| 191 | data->res.fattr = &data->fattr; |
| 192 | data->res.count = count; |
| 193 | data->res.eof = 0; |
Trond Myklebust | 0e574af | 2005-10-27 22:12:38 -0400 | [diff] [blame] | 194 | nfs_fattr_init(&data->fattr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 195 | |
Trond Myklebust | ec06c09 | 2006-03-20 13:44:27 -0500 | [diff] [blame] | 196 | /* Set up the initial task struct. */ |
Trond Myklebust | bdc7f02 | 2007-07-14 15:40:00 -0400 | [diff] [blame] | 197 | NFS_PROTO(inode)->read_setup(data, &msg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 198 | |
Chuck Lever | a3f565b | 2007-01-31 12:14:01 -0500 | [diff] [blame] | 199 | dprintk("NFS: %5u initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 200 | data->task.tk_pid, |
| 201 | inode->i_sb->s_id, |
| 202 | (long long)NFS_FILEID(inode), |
| 203 | count, |
| 204 | (unsigned long long)data->args.offset); |
Trond Myklebust | bdc7f02 | 2007-07-14 15:40:00 -0400 | [diff] [blame] | 205 | |
Trond Myklebust | 0773769 | 2007-10-25 18:42:54 -0400 | [diff] [blame] | 206 | task = rpc_run_task(&task_setup_data); |
| 207 | if (!IS_ERR(task)) |
| 208 | rpc_put_task(task); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 209 | } |
| 210 | |
| 211 | static void |
| 212 | nfs_async_read_error(struct list_head *head) |
| 213 | { |
| 214 | struct nfs_page *req; |
| 215 | |
| 216 | while (!list_empty(head)) { |
| 217 | req = nfs_list_entry(head->next); |
| 218 | nfs_list_remove_request(req); |
| 219 | SetPageError(req->wb_page); |
| 220 | nfs_readpage_release(req); |
| 221 | } |
| 222 | } |
| 223 | |
| 224 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 225 | * Generate multiple requests to fill a single page. |
| 226 | * |
| 227 | * We optimize to reduce the number of read operations on the wire. If we |
| 228 | * detect that we're reading a page, or an area of a page, that is past the |
| 229 | * end of file, we do not generate NFS read operations but just clear the |
| 230 | * parts of the page that would have come back zero from the server anyway. |
| 231 | * |
| 232 | * We rely on the cached value of i_size to make this determination; another |
| 233 | * client can fill pages on the server past our cached end-of-file, but we |
| 234 | * won't see the new data until our attribute cache is updated. This is more |
| 235 | * or less conventional NFS client behavior. |
| 236 | */ |
Trond Myklebust | 8d5658c | 2007-04-10 09:26:35 -0400 | [diff] [blame] | 237 | static int nfs_pagein_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 238 | { |
| 239 | struct nfs_page *req = nfs_list_entry(head->next); |
| 240 | struct page *page = req->wb_page; |
| 241 | struct nfs_read_data *data; |
Trond Myklebust | e9f7bee | 2006-09-08 09:48:54 -0700 | [diff] [blame] | 242 | size_t rsize = NFS_SERVER(inode)->rsize, nbytes; |
| 243 | unsigned int offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 244 | int requests = 0; |
| 245 | LIST_HEAD(list); |
| 246 | |
| 247 | nfs_list_remove_request(req); |
| 248 | |
Trond Myklebust | bcb71bb | 2007-04-02 18:48:28 -0400 | [diff] [blame] | 249 | nbytes = count; |
Trond Myklebust | e9f7bee | 2006-09-08 09:48:54 -0700 | [diff] [blame] | 250 | do { |
| 251 | size_t len = min(nbytes,rsize); |
| 252 | |
Trond Myklebust | 8d5658c | 2007-04-10 09:26:35 -0400 | [diff] [blame] | 253 | data = nfs_readdata_alloc(1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 254 | if (!data) |
| 255 | goto out_bad; |
| 256 | INIT_LIST_HEAD(&data->pages); |
| 257 | list_add(&data->pages, &list); |
| 258 | requests++; |
Trond Myklebust | e9f7bee | 2006-09-08 09:48:54 -0700 | [diff] [blame] | 259 | nbytes -= len; |
| 260 | } while(nbytes != 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 261 | atomic_set(&req->wb_complete, requests); |
| 262 | |
| 263 | ClearPageError(page); |
| 264 | offset = 0; |
Trond Myklebust | bcb71bb | 2007-04-02 18:48:28 -0400 | [diff] [blame] | 265 | nbytes = count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 266 | do { |
| 267 | data = list_entry(list.next, struct nfs_read_data, pages); |
| 268 | list_del_init(&data->pages); |
| 269 | |
| 270 | data->pagevec[0] = page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 271 | |
Trond Myklebust | bcb71bb | 2007-04-02 18:48:28 -0400 | [diff] [blame] | 272 | if (nbytes < rsize) |
| 273 | rsize = nbytes; |
| 274 | nfs_read_rpcsetup(req, data, &nfs_read_partial_ops, |
| 275 | rsize, offset); |
| 276 | offset += rsize; |
| 277 | nbytes -= rsize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 278 | } while (nbytes != 0); |
| 279 | |
| 280 | return 0; |
| 281 | |
| 282 | out_bad: |
| 283 | while (!list_empty(&list)) { |
| 284 | data = list_entry(list.next, struct nfs_read_data, pages); |
| 285 | list_del(&data->pages); |
| 286 | nfs_readdata_free(data); |
| 287 | } |
| 288 | SetPageError(page); |
| 289 | nfs_readpage_release(req); |
| 290 | return -ENOMEM; |
| 291 | } |
| 292 | |
Trond Myklebust | 8d5658c | 2007-04-10 09:26:35 -0400 | [diff] [blame] | 293 | static int nfs_pagein_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 294 | { |
| 295 | struct nfs_page *req; |
| 296 | struct page **pages; |
| 297 | struct nfs_read_data *data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 298 | |
Trond Myklebust | 8d5658c | 2007-04-10 09:26:35 -0400 | [diff] [blame] | 299 | data = nfs_readdata_alloc(npages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 300 | if (!data) |
| 301 | goto out_bad; |
| 302 | |
| 303 | INIT_LIST_HEAD(&data->pages); |
| 304 | pages = data->pagevec; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 305 | while (!list_empty(head)) { |
| 306 | req = nfs_list_entry(head->next); |
| 307 | nfs_list_remove_request(req); |
| 308 | nfs_list_add_request(req, &data->pages); |
| 309 | ClearPageError(req->wb_page); |
| 310 | *pages++ = req->wb_page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 311 | } |
| 312 | req = nfs_list_entry(data->pages.next); |
| 313 | |
Trond Myklebust | ec06c09 | 2006-03-20 13:44:27 -0500 | [diff] [blame] | 314 | nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 315 | return 0; |
| 316 | out_bad: |
| 317 | nfs_async_read_error(head); |
| 318 | return -ENOMEM; |
| 319 | } |
| 320 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 321 | /* |
Trond Myklebust | 0b67130 | 2006-11-14 16:12:23 -0500 | [diff] [blame] | 322 | * This is the callback from RPC telling us whether a reply was |
| 323 | * received or some error occurred (timeout or socket shutdown). |
| 324 | */ |
| 325 | int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data) |
| 326 | { |
| 327 | int status; |
| 328 | |
Chuck Lever | a3f565b | 2007-01-31 12:14:01 -0500 | [diff] [blame] | 329 | dprintk("NFS: %s: %5u, (status %d)\n", __FUNCTION__, task->tk_pid, |
Trond Myklebust | 0b67130 | 2006-11-14 16:12:23 -0500 | [diff] [blame] | 330 | task->tk_status); |
| 331 | |
| 332 | status = NFS_PROTO(data->inode)->read_done(task, data); |
| 333 | if (status != 0) |
| 334 | return status; |
| 335 | |
| 336 | nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count); |
| 337 | |
| 338 | if (task->tk_status == -ESTALE) { |
Benny Halevy | 3a10c30 | 2008-01-23 08:58:59 +0200 | [diff] [blame] | 339 | set_bit(NFS_INO_STALE, &NFS_I(data->inode)->flags); |
Trond Myklebust | 0b67130 | 2006-11-14 16:12:23 -0500 | [diff] [blame] | 340 | nfs_mark_for_revalidate(data->inode); |
| 341 | } |
Trond Myklebust | 0b67130 | 2006-11-14 16:12:23 -0500 | [diff] [blame] | 342 | return 0; |
| 343 | } |
| 344 | |
| 345 | static int nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data) |
| 346 | { |
| 347 | struct nfs_readargs *argp = &data->args; |
| 348 | struct nfs_readres *resp = &data->res; |
| 349 | |
| 350 | if (resp->eof || resp->count == argp->count) |
| 351 | return 0; |
| 352 | |
| 353 | /* This is a short read! */ |
| 354 | nfs_inc_stats(data->inode, NFSIOS_SHORTREAD); |
| 355 | /* Has the server at least made some progress? */ |
| 356 | if (resp->count == 0) |
| 357 | return 0; |
| 358 | |
| 359 | /* Yes, so retry the read at the end of the data */ |
| 360 | argp->offset += resp->count; |
| 361 | argp->pgbase += resp->count; |
| 362 | argp->count -= resp->count; |
| 363 | rpc_restart_call(task); |
| 364 | return -EAGAIN; |
| 365 | } |
| 366 | |
| 367 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | * Handle a read reply that fills part of a page. |
| 369 | */ |
Trond Myklebust | ec06c09 | 2006-03-20 13:44:27 -0500 | [diff] [blame] | 370 | static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 371 | { |
Trond Myklebust | ec06c09 | 2006-03-20 13:44:27 -0500 | [diff] [blame] | 372 | struct nfs_read_data *data = calldata; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 373 | struct nfs_page *req = data->req; |
| 374 | struct page *page = req->wb_page; |
| 375 | |
Trond Myklebust | ec06c09 | 2006-03-20 13:44:27 -0500 | [diff] [blame] | 376 | if (nfs_readpage_result(task, data) != 0) |
| 377 | return; |
Trond Myklebust | 0b67130 | 2006-11-14 16:12:23 -0500 | [diff] [blame] | 378 | |
| 379 | if (likely(task->tk_status >= 0)) { |
| 380 | nfs_readpage_truncate_uninitialised_page(data); |
| 381 | if (nfs_readpage_retry(task, data) != 0) |
| 382 | return; |
| 383 | } |
| 384 | if (unlikely(task->tk_status < 0)) |
| 385 | SetPageError(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 386 | if (atomic_dec_and_test(&req->wb_complete)) { |
| 387 | if (!PageError(page)) |
| 388 | SetPageUptodate(page); |
| 389 | nfs_readpage_release(req); |
| 390 | } |
| 391 | } |
| 392 | |
Trond Myklebust | ec06c09 | 2006-03-20 13:44:27 -0500 | [diff] [blame] | 393 | static const struct rpc_call_ops nfs_read_partial_ops = { |
| 394 | .rpc_call_done = nfs_readpage_result_partial, |
| 395 | .rpc_release = nfs_readdata_release, |
| 396 | }; |
| 397 | |
Trond Myklebust | 1de3fc1 | 2006-05-25 01:40:44 -0400 | [diff] [blame] | 398 | static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data) |
| 399 | { |
| 400 | unsigned int count = data->res.count; |
| 401 | unsigned int base = data->args.pgbase; |
| 402 | struct page **pages; |
| 403 | |
Trond Myklebust | 79558f3 | 2006-08-22 13:44:32 -0400 | [diff] [blame] | 404 | if (data->res.eof) |
| 405 | count = data->args.count; |
Trond Myklebust | 1de3fc1 | 2006-05-25 01:40:44 -0400 | [diff] [blame] | 406 | if (unlikely(count == 0)) |
| 407 | return; |
| 408 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; |
| 409 | base &= ~PAGE_CACHE_MASK; |
| 410 | count += base; |
| 411 | for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++) |
| 412 | SetPageUptodate(*pages); |
Trond Myklebust | 0b67130 | 2006-11-14 16:12:23 -0500 | [diff] [blame] | 413 | if (count == 0) |
| 414 | return; |
| 415 | /* Was this a short read? */ |
| 416 | if (data->res.eof || data->res.count == data->args.count) |
Trond Myklebust | 1de3fc1 | 2006-05-25 01:40:44 -0400 | [diff] [blame] | 417 | SetPageUptodate(*pages); |
| 418 | } |
| 419 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 420 | /* |
| 421 | * This is the callback from RPC telling us whether a reply was |
| 422 | * received or some error occurred (timeout or socket shutdown). |
| 423 | */ |
Trond Myklebust | ec06c09 | 2006-03-20 13:44:27 -0500 | [diff] [blame] | 424 | static void nfs_readpage_result_full(struct rpc_task *task, void *calldata) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 425 | { |
Trond Myklebust | ec06c09 | 2006-03-20 13:44:27 -0500 | [diff] [blame] | 426 | struct nfs_read_data *data = calldata; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 427 | |
Trond Myklebust | 0b67130 | 2006-11-14 16:12:23 -0500 | [diff] [blame] | 428 | if (nfs_readpage_result(task, data) != 0) |
| 429 | return; |
Trond Myklebust | 1de3fc1 | 2006-05-25 01:40:44 -0400 | [diff] [blame] | 430 | /* |
Trond Myklebust | 0b67130 | 2006-11-14 16:12:23 -0500 | [diff] [blame] | 431 | * Note: nfs_readpage_retry may change the values of |
Trond Myklebust | 1de3fc1 | 2006-05-25 01:40:44 -0400 | [diff] [blame] | 432 | * data->args. In the multi-page case, we therefore need |
Trond Myklebust | 0b67130 | 2006-11-14 16:12:23 -0500 | [diff] [blame] | 433 | * to ensure that we call nfs_readpage_set_pages_uptodate() |
| 434 | * first. |
Trond Myklebust | 1de3fc1 | 2006-05-25 01:40:44 -0400 | [diff] [blame] | 435 | */ |
| 436 | if (likely(task->tk_status >= 0)) { |
| 437 | nfs_readpage_truncate_uninitialised_page(data); |
| 438 | nfs_readpage_set_pages_uptodate(data); |
Trond Myklebust | 0b67130 | 2006-11-14 16:12:23 -0500 | [diff] [blame] | 439 | if (nfs_readpage_retry(task, data) != 0) |
| 440 | return; |
| 441 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 442 | while (!list_empty(&data->pages)) { |
| 443 | struct nfs_page *req = nfs_list_entry(data->pages.next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 444 | |
Trond Myklebust | 1de3fc1 | 2006-05-25 01:40:44 -0400 | [diff] [blame] | 445 | nfs_list_remove_request(req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 446 | nfs_readpage_release(req); |
| 447 | } |
| 448 | } |
| 449 | |
Trond Myklebust | ec06c09 | 2006-03-20 13:44:27 -0500 | [diff] [blame] | 450 | static const struct rpc_call_ops nfs_read_full_ops = { |
| 451 | .rpc_call_done = nfs_readpage_result_full, |
| 452 | .rpc_release = nfs_readdata_release, |
| 453 | }; |
| 454 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 455 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 456 | * Read a page over NFS. |
| 457 | * We read the page synchronously in the following case: |
| 458 | * - The error flag is set for this page. This happens only when a |
| 459 | * previous async read operation failed. |
| 460 | */ |
| 461 | int nfs_readpage(struct file *file, struct page *page) |
| 462 | { |
| 463 | struct nfs_open_context *ctx; |
| 464 | struct inode *inode = page->mapping->host; |
| 465 | int error; |
| 466 | |
| 467 | dprintk("NFS: nfs_readpage (%p %ld@%lu)\n", |
| 468 | page, PAGE_CACHE_SIZE, page->index); |
Chuck Lever | 91d5b47 | 2006-03-20 13:44:14 -0500 | [diff] [blame] | 469 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGE); |
| 470 | nfs_add_stats(inode, NFSIOS_READPAGES, 1); |
| 471 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 472 | /* |
| 473 | * Try to flush any pending writes to the file.. |
| 474 | * |
| 475 | * NOTE! Because we own the page lock, there cannot |
| 476 | * be any new pending writes generated at this point |
| 477 | * for this page (other pages can be written to). |
| 478 | */ |
| 479 | error = nfs_wb_page(inode, page); |
| 480 | if (error) |
Trond Myklebust | de05a0c | 2007-05-20 13:05:05 -0400 | [diff] [blame] | 481 | goto out_unlock; |
| 482 | if (PageUptodate(page)) |
| 483 | goto out_unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 484 | |
Trond Myklebust | 5f004cf | 2006-09-14 14:03:14 -0400 | [diff] [blame] | 485 | error = -ESTALE; |
| 486 | if (NFS_STALE(inode)) |
Trond Myklebust | de05a0c | 2007-05-20 13:05:05 -0400 | [diff] [blame] | 487 | goto out_unlock; |
Trond Myklebust | 5f004cf | 2006-09-14 14:03:14 -0400 | [diff] [blame] | 488 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 489 | if (file == NULL) { |
Trond Myklebust | cf1308f | 2006-11-19 16:44:52 -0500 | [diff] [blame] | 490 | error = -EBADF; |
Trond Myklebust | d530838 | 2005-11-04 15:33:38 -0500 | [diff] [blame] | 491 | ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 492 | if (ctx == NULL) |
Trond Myklebust | de05a0c | 2007-05-20 13:05:05 -0400 | [diff] [blame] | 493 | goto out_unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 494 | } else |
Trond Myklebust | cd3758e | 2007-08-10 17:44:32 -0400 | [diff] [blame] | 495 | ctx = get_nfs_open_context(nfs_file_open_context(file)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 496 | |
Trond Myklebust | 8e0969f | 2006-12-13 15:23:44 -0500 | [diff] [blame] | 497 | error = nfs_readpage_async(ctx, inode, page); |
| 498 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 499 | put_nfs_open_context(ctx); |
| 500 | return error; |
Trond Myklebust | de05a0c | 2007-05-20 13:05:05 -0400 | [diff] [blame] | 501 | out_unlock: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 502 | unlock_page(page); |
| 503 | return error; |
| 504 | } |
| 505 | |
| 506 | struct nfs_readdesc { |
Trond Myklebust | 8b09bee | 2007-04-02 18:48:28 -0400 | [diff] [blame] | 507 | struct nfs_pageio_descriptor *pgio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 508 | struct nfs_open_context *ctx; |
| 509 | }; |
| 510 | |
| 511 | static int |
| 512 | readpage_async_filler(void *data, struct page *page) |
| 513 | { |
| 514 | struct nfs_readdesc *desc = (struct nfs_readdesc *)data; |
| 515 | struct inode *inode = page->mapping->host; |
| 516 | struct nfs_page *new; |
| 517 | unsigned int len; |
Trond Myklebust | de05a0c | 2007-05-20 13:05:05 -0400 | [diff] [blame] | 518 | int error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 519 | |
Trond Myklebust | de05a0c | 2007-05-20 13:05:05 -0400 | [diff] [blame] | 520 | error = nfs_wb_page(inode, page); |
| 521 | if (error) |
| 522 | goto out_unlock; |
| 523 | if (PageUptodate(page)) |
| 524 | goto out_unlock; |
| 525 | |
Trond Myklebust | 49a70f2 | 2006-12-05 00:35:38 -0500 | [diff] [blame] | 526 | len = nfs_page_length(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 527 | if (len == 0) |
| 528 | return nfs_return_empty_page(page); |
Trond Myklebust | de05a0c | 2007-05-20 13:05:05 -0400 | [diff] [blame] | 529 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 530 | new = nfs_create_request(desc->ctx, inode, page, 0, len); |
Trond Myklebust | de05a0c | 2007-05-20 13:05:05 -0400 | [diff] [blame] | 531 | if (IS_ERR(new)) |
| 532 | goto out_error; |
| 533 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 534 | if (len < PAGE_CACHE_SIZE) |
Christoph Lameter | eebd2aa | 2008-02-04 22:28:29 -0800 | [diff] [blame] | 535 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
Fred Isaman | f8512ad | 2008-03-19 11:24:39 -0400 | [diff] [blame^] | 536 | if (!nfs_pageio_add_request(desc->pgio, new)) { |
| 537 | error = desc->pgio->pg_error; |
| 538 | goto out_unlock; |
| 539 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 540 | return 0; |
Trond Myklebust | de05a0c | 2007-05-20 13:05:05 -0400 | [diff] [blame] | 541 | out_error: |
| 542 | error = PTR_ERR(new); |
| 543 | SetPageError(page); |
| 544 | out_unlock: |
| 545 | unlock_page(page); |
| 546 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 547 | } |
| 548 | |
| 549 | int nfs_readpages(struct file *filp, struct address_space *mapping, |
| 550 | struct list_head *pages, unsigned nr_pages) |
| 551 | { |
Trond Myklebust | 8b09bee | 2007-04-02 18:48:28 -0400 | [diff] [blame] | 552 | struct nfs_pageio_descriptor pgio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 553 | struct nfs_readdesc desc = { |
Trond Myklebust | 8b09bee | 2007-04-02 18:48:28 -0400 | [diff] [blame] | 554 | .pgio = &pgio, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 555 | }; |
| 556 | struct inode *inode = mapping->host; |
| 557 | struct nfs_server *server = NFS_SERVER(inode); |
Trond Myklebust | 8b09bee | 2007-04-02 18:48:28 -0400 | [diff] [blame] | 558 | size_t rsize = server->rsize; |
| 559 | unsigned long npages; |
Trond Myklebust | 5f004cf | 2006-09-14 14:03:14 -0400 | [diff] [blame] | 560 | int ret = -ESTALE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 561 | |
| 562 | dprintk("NFS: nfs_readpages (%s/%Ld %d)\n", |
| 563 | inode->i_sb->s_id, |
| 564 | (long long)NFS_FILEID(inode), |
| 565 | nr_pages); |
Chuck Lever | 91d5b47 | 2006-03-20 13:44:14 -0500 | [diff] [blame] | 566 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 567 | |
Trond Myklebust | 5f004cf | 2006-09-14 14:03:14 -0400 | [diff] [blame] | 568 | if (NFS_STALE(inode)) |
| 569 | goto out; |
| 570 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 571 | if (filp == NULL) { |
Trond Myklebust | d530838 | 2005-11-04 15:33:38 -0500 | [diff] [blame] | 572 | desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 573 | if (desc.ctx == NULL) |
| 574 | return -EBADF; |
| 575 | } else |
Trond Myklebust | cd3758e | 2007-08-10 17:44:32 -0400 | [diff] [blame] | 576 | desc.ctx = get_nfs_open_context(nfs_file_open_context(filp)); |
Trond Myklebust | 8b09bee | 2007-04-02 18:48:28 -0400 | [diff] [blame] | 577 | if (rsize < PAGE_CACHE_SIZE) |
| 578 | nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0); |
| 579 | else |
| 580 | nfs_pageio_init(&pgio, inode, nfs_pagein_one, rsize, 0); |
| 581 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 582 | ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc); |
Trond Myklebust | 8b09bee | 2007-04-02 18:48:28 -0400 | [diff] [blame] | 583 | |
| 584 | nfs_pageio_complete(&pgio); |
| 585 | npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
| 586 | nfs_add_stats(inode, NFSIOS_READPAGES, npages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 587 | put_nfs_open_context(desc.ctx); |
Trond Myklebust | 5f004cf | 2006-09-14 14:03:14 -0400 | [diff] [blame] | 588 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 589 | return ret; |
| 590 | } |
| 591 | |
David Howells | f7b422b | 2006-06-09 09:34:33 -0400 | [diff] [blame] | 592 | int __init nfs_init_readpagecache(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 593 | { |
| 594 | nfs_rdata_cachep = kmem_cache_create("nfs_read_data", |
| 595 | sizeof(struct nfs_read_data), |
| 596 | 0, SLAB_HWCACHE_ALIGN, |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 597 | NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 598 | if (nfs_rdata_cachep == NULL) |
| 599 | return -ENOMEM; |
| 600 | |
Matthew Dobson | 93d2341 | 2006-03-26 01:37:50 -0800 | [diff] [blame] | 601 | nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ, |
| 602 | nfs_rdata_cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 603 | if (nfs_rdata_mempool == NULL) |
| 604 | return -ENOMEM; |
| 605 | |
| 606 | return 0; |
| 607 | } |
| 608 | |
David Brownell | 266bee8 | 2006-06-27 12:59:15 -0700 | [diff] [blame] | 609 | void nfs_destroy_readpagecache(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 610 | { |
| 611 | mempool_destroy(nfs_rdata_mempool); |
Alexey Dobriyan | 1a1d92c | 2006-09-27 01:49:40 -0700 | [diff] [blame] | 612 | kmem_cache_destroy(nfs_rdata_cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 613 | } |